The hype of artificial intelligence is running up against pricey realities, at least for now. MacBooks and Xboxes are getting more expensive due to “AI inflation.” Central bankers are warning the AI boom could soon trigger a financial crash. Ford, meanwhile, has hired hundreds of engineers to do the work that artificial intelligence software could not. It has opened debate as to whether the benefits of AI are worth the costs.

What did the commentators say?

The mass buildout of data centers has created the “global shortage of memory and storage chips” behind the new device price hikes, said CBS News. MacBook Pros are going from $1,699 to $1,999, while the entry-level Xbox is rising to $499 from $399. The demand from tech giants such as Alphabet, Amazon, Meta and Oracle leaves fewer chips for “regular consumer devices,” Wedbush Securities’ Dan Ives said to the outlet. “That just further drives up prices.”

That inflation is typical of any “technological revolution,” Jennifer Schonberger said at Yahoo Finance. Big investment in new tech puts “pressures on resources with a lot of demand chasing a limited supply.” For now, AI inflation is “screwing with the rest of the economy,” said John Herrman at New York magazine. Once-accessible goods “seem to be slipping out of reach” of ordinary Americans, which “could meaningfully contribute to an already apocalyptic” national mood.

But the data center surge could come to a sudden, thudding halt if those big companies do not soon see a return on their investment. The Bank for International Settlements (BIS) on Sunday warned that an end to the data center “spending spree” could “rattle financial markets and damage the global economy,” said the Financial Times. There are “instructive parallels” in the dotcom boom of the 1990s and the buildout of British railways in the 19th century, the BIS said in its annual economic report. “These episodes ended with an eventual reversal in investment, inducing economy-wide recessions.”

The payoffs are fuzzy. Ford has hired 350 “veteran engineers” to “reprogram the artificial intelligence tools that weren’t getting the job done,” said Bloomberg. AI is a “fantastic tool,” Ford executive Charles Poon said to reporters last week, per the outlet. But it is “only as good as the information you use to train it.” Ford’s dramatic shift to AI was a “grave mistake” that affected the company’s quality control, said Danni Santana at Moneywise. That happens “when you overcommit to the technology without properly training it.” The company will still use AI, but its experience demonstrates that the technology cannot “completely replace humans in the workforce” for now.

What next?

Businesses that enthusiastically embraced artificial intelligence are now trying to find a balance. Many firms urged their employees to “integrate AI tools into their work” only to see their “AI spending bills double or triple,” said The Wall Street Journal. Executives at enterprises like Uber, Meta and Microsoft are now looking to “steer workers toward cheaper, homegrown tools” and help those employees “hone their skills.”

If a hacker steals your password, you can create a new one, but if someone gains access to your fingerprint or iris data, you can hardly replace your fingers or eyes. But a new study has shown promise with a technique that allows users to “update” their fingerprints, which could make us all safer online.

Spy novels

Concern about the security of using fingerprints instead of passwords has grown this month amid reports that scammers could extract close-ups of fingerprints from social media photos and “enhance them with AI”, said Moneywise. The criminals could then use the victim’s unique fingerprint ID to gain access to their accounts, or launch identity theft and phishing attacks, although they would still need access to a physical scanner, like a smartphone unlock key, to use the cloned fingerprint.

It “sounds like the stuff out of spy novels or ‘Mission Impossible’”, Vyas Sekar, a professor at Carnegie Mellon University, told CBS News, but “in theory, it’s possible, especially if people are posting high-resolution images”. In 2014, a hacker claimed to have cloned a fingerprint of European Commission President Ursula von der Leyen, then Germany’s defence minister, using close-up photos taken at a press event.

‘Scrambled and compressed’

A study in the International Journal of Computational Vision and Robotics has found that “irreversible identity theft” can be “largely avoided” by giving users a chance to “reset” fingerprints and other biometrics, said TechXplore.

The method is “similar to changing a password”, said Knowridge. Rather than storing a person’s original fingerprint or other biometric information directly, it transforms their data into a protected version. To do this, it identifies unique features in a fingerprint image, such as distinctive patterns and points, and “uses mathematical methods to convert these features into a different form that is difficult to reverse-engineer”. The data is then “further scrambled and compressed” into a secure digital version.

In this form, it can still verify a person’s identity, but the original biometric data is hidden. If the protected version is ever compromised, it can be “cancelled and replaced”. Even if hackers gained access to the stored information, the user would not be permanently exposed.

For years, philosophy graduates have been the “butt of jokes about unemployable degrees”, said Thibault Spirlet in Business Insider. Now, they can earn six-figure salaries as the “world’s most powerful AI companies” try to “shape how machines think and behave”.

High-profile philosophers are already “embedded” in top AI firms. Amanda Askell is resident philosopher at Anthropic, and Iason Gabriel and Henry Shevlin work at Google DeepMind. OpenAI’s Sam Altman also claimed that the company employed “hundreds of moral philosophers” when designing rules for ChatGPT. But there is rising suspicion that there are ulterior motives at play.

‘Arc of redemption’

“Unemployed coders take note,” said The Economist: “there seems to be no shortage of work for philosophers of AI.” There are “thorny problems” in the developing field – “a philosopher’s favourite sort”.

Some “ancient” philosophical considerations are at the core of the contemporary tech industry. The idea of “Socratic ignorance” – that wisdom is an individual realising the extent of what they do not know – is a major principle in AI development used to avoid “sycophancy”.

Deliberating whether a system should follow deontological aims (“strict rules” against “lying, coercion and treating people as a means rather than an end”), or consequentialist ones (which weigh “costs against benefits”) is also a common dilemma for developers.

Philosophy is key to safety practices, too. Implementing the concept of “AI constitutionalism” – where legally or morally authoritative texts are used as a base of “scaffolding” to direct the system – aims to prevent “ominous behaviour” from the models.

Anthropic revealed earlier this year that its Claude constitution included sources as “diverse as Immanuel Kant, Apple’s terms of service and the Universal Declaration of Human Rights”. This has been nicknamed the company’s “soul doc”.

A rise in demand for philosophers has also coincided with a decline in admissions for computer science students, said Lance Eliot in Forbes. Arguably, computer science has become a “dead-end endeavour”, creating “automation that replaces the humans who made it all possible”. AI programming once held the “promise of big bucks and a stellar career”. This may just be a minor “course correction”, as no doubt degrees that directly relate to AI will remain important, but nonetheless, philosophy is experiencing an “amazing arc of redemption”.

But influence goes both ways and is “not limited to Silicon Valley”, said Alexandra Tremayne-Pengelly on Observer. Philosophy is impacting tech, but the demands of the AI industry are reshaping the “long-standing” landscape of philosophical thought. Academia is “rapidly adapting” as foundational questions regarding consciousness, morality, minds and computation have taken on a “new urgency”.

Suspicion and ‘ethics-washing’

The two disciplines of computer science and philosophy have “never been quite as entangled” nor as “fraught” as they are now, said Lila Shroff in The Atlantic. In a fundamental sense, the “careful thought” of philosophy is “at odds with the frenetic pace of AI”. In turn, some experts are concerned that “misaligned incentives” will encourage a “rush of low-quality work”.

There is a “degree of suspicion” in the academic world about philosophers migrating to AI firms, said Joel Khalili in Wired. The whole industry poses significant ethical risks. These programmes could be used to “develop new weapons of mass destruction, undermine democracy, or entrench existing social iniquities”.

But the greatest fear is of “ethics-washing”. Hiring philosophers to train systems not only demonstrates to the public that these models are so advanced that they warrant the attention of “serious people”, but also shows that companies are “outwardly performing a commitment to AI safety”.

In a broader sense, there are growing fears that philosophical research is becoming an “extension of the marketing function” of labs. And even if philosophers are given “free rein” in tech companies, ultimately, they are “accountable to investors and shareholders”. Essentially, “if a for-profit AI company signs your pay cheque, might that compromise your research?”

It’s funny what the pressure of a ticking clock can do to a prime minister, said Hannah Barnes in The New Statesman. Just a few months ago, when more than 60 Labour MPs signed an open letter calling for a social media ban for under-16s, Keir Starmer wasn’t convinced. His own teenage children, he told MPs, had benefitted from using social media.

But with a leadership challenge looming, the panicking PM has suddenly “latched” on to this popular policy. He announced this week that, from next year, under-16s would be banned from social media platforms including Snapchat, TikTok, YouTube, X and Instagram. “I am not prepared to compromise on the safety and happiness of our children,” he declared.

Fencing the ocean

Strong rhetoric, said Christopher Snowdon in The Critic, yet this policy “has all the hallmarks of another government failure”. In Australia, a similar ban has proven almost totally ineffectual. Some 70% of parents whose children used social media before the ban say they still do so. Australia's e-safety commissioner, who is in charge of enforcing the ban, has said that it's like trying to “fence the ocean”.

There are countless reasons why blanket bans don’t work, said Chris Stokel-Walker in The Independent. You have to outwit 14-year-olds armed with VPNs, older siblings, and “group chats full of instructions”. Age verification is difficult. And even if you do get the kids off big social media sites, they “don’t suddenly take up whittling”. They move to darker, harder-to-regulate parts of the web: to Discord servers, private groups. “A serious government would force platforms to change their products”, as the campaigner Ian Russell has urged, so that they don’t pummel children with harmful content. “But this is a dead-duck administration grasping for a legacy.”

‘Multi-front’ battle

Besides, whatever happened to parenting, asked Daniel Hannan in the Daily Mail. Why use the “heavy hand” of the state, when the tide is already turning and most people now limit their kids’ screen time?

If only it were that easy, said Sam Leith in The Spectator. Like millions of other exhausted parents, I spend every day waging a “multi-front” battle against the screens: the iPhone and iPad, computer and PlayStation. I’ve set time limits, banned sites at the router, used third-party blockers. And yet still my children’s brains are being rotted by “AI slop, vacuous influencers” and silly videos, while I worry about far more damaging content. So, even if I have doubts about how it will work, I love the idea of the ban – like nine out of ten British parents consulted.

Until now, we’ve been fighting against the world’s most powerful tech companies almost totally unarmed, said Isabel Oakeshott in The Daily Telegraph. At last, the PM is giving us a “half-decent” weapon. Of course some children will get round the ban, as they do with underage drinking. But we haven’t abandoned the law on that, have we? I cannot wait to tell my teens “that lounging around on their screens is ‘literally illegal’”.

The government is pressing ahead with an AI tool used to check the ages of migrant children despite warnings from its own advisers that it is “hideously inaccurate”, said The Independent.

The Home Office announced last month that facial age estimation (FAE) technology would be used by immigration officers from 2027 to “crack down on fake claims by small boat arrivals posing as children”, but two reports have identified serious problems with it.

How does FAE work?

AI-powered FAE tools have been used to prevent children from accessing age-restricted goods and services, including cigarettes, alcohol and adult-only online content.

Now, the “hardening politics around asylum” has opened a new market with “far higher stakes” – using AI to determine whether undocumented migrants are under or over 18, said Lighthouse Reports, in collaboration with The Independent and Wired.

A photograph is fed into an AI system and goes through several layers of analysis, each picking up “increasingly subtle patterns”, said Oli Buckley, a cybersecurity professor at Loughborough University, on The Conversation.

The system, which is “trained on millions of photographs of people whose ages are already known”, has learned to associate patterns in a face with likely age ranges. It studies skin texture, the depth of lines around the eyes, bone structure and the distribution of soft tissue, before delivering a “probability distribution”, which is closer to “most likely between 17 and 21” than “this person is 18.”

All of this matters because at the UK border, deciding whether someone is 17 or 19 is a very “consequential judgment”. If you “get it wrong one way”, a “vulnerable child” loses legal protections they’re “entitled to”, but “if it’s wrong in the other direction”, then an adult “enters a system designed for minors”.

What are the issues?

The technology misclassified more than a third of 16-year-olds as adults and in some tests it was shown to give the wrong assessment in 70% of cases, according to an audit by Lighthouse Reports.

A separate, leaked report, which the Home Office tried to withhold, found the technology is least accurate when trying to assess migrants from countries such as Eritrea and Sudan, which have the highest number of small boat migrants arriving in the UK. This has led to accusations that the technology has “baked-in racial bias”, said The Independent.

The report also warned that “error rates are particularly high for female child migrants from Sub-Saharan Africa”. Estimations were out by 4.6 years on average – “meaning a 14-year-old girl could be predicted to be an adult”. The AI tool could also be less accurate for people with visible ageing caused by “stress of travel”.

Will it still be used?

The Home Office said it has “rigorous processes in place to verify an individual’s age” and the “groundbreaking assistive tool is designed as an additional source of information for immigration officers, and does not replace or overrule human judgment”.

But Anna Bacciarelli, senior researcher on technology, rights and investigations at Human Rights Watch, told Lighthouse Reports that this was of little reassurance because of the “well-established phenomenon” of automation bias, where people tend to trust a computer’s decision over their own judgment.

The UK’s use of the technology sets a “dangerous precedent”, she said, and “other countries at entry points across Europe are likely to follow suit”. As a result, “use of this inaccurate and invasive technology could become widespread”.

Apple introduced the iPhone to the world in 2007. That was the same year that birth rates around the world began to decline. And the two developments may be related.

“If your sex life is dead, you can blame Steve Jobs,” Brandon Vigliarolo said at The Register. Two new studies suggest smartphones are responsible for the baby bust. One study found the iPhone “caused as much as half of the fertility decline” from 2007 to 2011, said The New York Times. A second study of 128 countries found that teen pregnancies declined “once smartphones became a mass phenomenon.” It may be that people “began to socialize more on their phones and less in person,” or it could be that the technology “made pornography more accessible.”

Experts suggested caution is needed. Smartphones are just one “example of the kinds of social influences” that may have reduced fertility, said Wellesley College’s Phillip B. Levine to the Times.

‘Awkward, antisocial puppies’

Phones have “turned us into awkward, antisocial puppies who can’t handle eye contact,” said Lauren Veldhuizen at the National Review. The rise of smartphone technology has thus created a world in which “fewer people date and fewer babies are born.” Some might see the decline of teen pregnancies, in particular, as a positive development. But that would be true only if the decline were the result of an “increasing respect for the purpose of sex within the confines of marriage” instead of our increasing “inability to speak to one another.”

The media has glommed onto the new studies because of a collective mood of “total paranoia and doom about smartphones,” said Elizabeth Nolan Brown at Reason. The biggest plunges in the 2007-2011 study were among 15- to 24-year-old females, suggesting more girls and women are “avoiding unintended pregnancy at young ages.” The study’s time frame might also simply reflect the impact of the Great Recession. The research should be greeted “with some skepticism.”

Smartphones “short-circuit the deep-seated human need to have your kids keep you company,” said Noah Smith at Noahpinion. We are choosing to “forsake each other’s company to stare eternally into a black mirror.”

‘No easy fix’

Maybe smartphones first tarnished dating, but AI “might finish the job,” said Eric Levitz at Vox. Streaming and social media have helped us isolate from each other, yet online platforms could not discuss “your career anxieties, favorite Civil War battle or debilitating fear of iguanas.” Claude, ChatGPT, Gemini and other artificial intelligence chatbots can. “Humanity may be scrolling its way out of existence.”

There is “no easy fix here,” said Axios. Politicians have proposed “baby bonuses, tax credits or better child care and parental leave policies” to solve the fertility crisis, all to no avail. “Perhaps the solution is that everyone toss their phones into the sea.”

What happened

Fox Corp. said Monday it was buying streaming and smart-TV company Roku for $22 billion, its first major acquisition since chief executive Lachlan Murdoch cemented control of his family’s media empire last year. The deal will give Fox, with its news and live sports content, a foothold in the more than 100 million households that use Roku’s platform.

Who said what

The cash-and-stock deal “would make the Murdoch media empire a formidable contender in the streaming wars,” positioning Fox to “reach customers who are abandoning traditional TV,” The New York Times said. Specifically, it would transform the company into a “major player in free, ad-supported streaming,” The Washington Post said, combining Fox-owned Tubi with Roku’s own “free-to-stream, ad-supported offering.”

Fox’s “bigger play here is advertising revenue, something all the major streamers are now jockeying for,” Forrester research director Mike Proulx said in a statement. “If this deal closes, Fox will control more of what viewers watch, how they discover it and how it gets monetized.”

What next?

Fox and Roku said their merger, expected to close in the first half of 2027, would create the “third-largest player in U.S. television by share of viewing.”

Another major data centre has been given the green light in the UK, further cementing the country’s status as Europe’s AI front-runner. The government has approved a data centre on a huge green-belt site in Slough, Berkshire, despite claims it could derail the project to build Heathrow’s third runway.

The company had appealed after the council refused to rule on the project, which it said would sit in “one of the most fragile and vulnerable parts of the green belt around London”. Since coming to power, Labour has “repeatedly bypassed local authorities to support data centre developments”, said The Telegraph. Matthew Pennycook, the housing minister, said that there was a “continuing and unprecedented demand” for such projects.

How many data centres are there in the UK?

The UK is at the forefront of Europe’s data centre roll-out; it hosts 523 out of the continent’s 2,269 data centres as of last year, said Euronews. It is “striking” that China (home to 449 centres), “despite its strength as a technology and innovation power”, ranks behind the UK, as well as Germany (529 centres).

However, all three are dwarfed by the US, which last year boasted 5,427 data centres. The only other countries with more than 300 centres are Canada, France and Australia.

What’s the latest?

A multibillion-pound AI data centre in Wales was jeopardised by “ministerial dithering”, said The Telegraph. British data centre company Era4 said it had secured permission and financing for the project at the former Liberty Steel works in Newport, but that the “project had faced months of delays” because Kanishka Narayan, the AI minister, “failed to push through permission for it to access power from a nearby battery plant”.

Era4 said the Department for Science, Innovation and Technology had given “no indication of when a decision would be made or that the project would be approved”. Tom Humphreys, Era4’s chief executive, said the company was looking at sites in Europe as an alternative.

Other tech companies have also “complained of a struggle to build AI infrastructure in Britain”, said the paper. OpenAI recently announced it was pausing work on a data centre in the north of England due to high energy costs.

What’s planned for the future?

The number of data centres in the UK is set to increase by almost a fifth over the next five years, said the BBC last year, when there were already an estimated 477. Work on the biggest, a £10 billion data centre near Newcastle for US wealth management firm Blackstone Group, is due to begin in 2031. It will involve “10 giant buildings” covering more than half a million square metres – “the size of several large shopping centres”.

The majority will be built in London and neighbouring counties, despite “concerns about the huge amount of energy and water” they’ll consume, as well as the “potential knock-on effect” on domestic energy bills.

How much environmental impact do they have?

Officials recently admitted that Britain’s data centre boom could “draw 40% more electricity than thought a few months ago”, said The Times. More than 100 data centres are seeking grid connections for 50 gigawatts of electricity capacity – “more than the whole of Britain’s peak demand on a typical day”. MPs are calling for a “national conversation on the environmental impacts”.

“The previous projections were already unfathomable,” said Oliver Hayes, head of policy and campaigns at environmental charity Global Action Plan. Adding 40% on top is absurd.”

There are also concerns about the burning of fossil fuels to meet power demand, potentially jeopardising climate goals. There has been a “marked shift over the past year in willingness of UK developers – and authorities – to consider using fossil fuels to power the UK’s AI ambitions”, said The Guardian. More than 100 new UK data centres plan to burn gas, said “some potentially doing so permanently”.

Under-16s in the UK will be banned from social media under radical new plans set out by the prime minister today.

In a televised speech in Downing Street, Keir Starmer said he was “calling time on a system that’s failing our kids”. And while this was not a “cost-free decision”, governing “is always about choices, and it’s clear to me that a full ban is the right choice”.

Polling by YouGov suggests broad public support for the decision, with 77% of parents backing a ban. But parents were also split on whether a ban would work, with 45% of those surveyed saying it would be effective and 46% disagreeing.

How will it work?

The UK ban will cover the most popular social media platforms, including Snapchat, TikTok, YouTube, Instagram, Facebook and X (formerly Twitter), but not encrypted messaging apps like WhatsApp and Signal.

The government says it will “go further than any other country”, with its policy also including blocks on live-streaming and stranger communication for under-16s. Gaming sites will be impacted and the government is also looking at overnight curfews and breaks in infinite scrolling for 16- to 18-year-olds. A minimum age of 18 will be enforced on “romantic companion” AI chatbots, designed to simulate sexual relationships or roleplay with users.

As ever the devil will be in the detail. The government has said new restrictions will be enforced through “highly effective age assurance” systems, including facial age estimation using digital cameras. The media regulator Ofcom “will conduct a rapid study on what is effective age assurance for verifying whether someone is over 16”, said the government’s official announcement.

The PM said he hopes to pass the necessary legislation by Christmas, with the ban coming into effect in spring 2027.

Will it work?

The government has been accused of rushing out plans for a social media ban “without considering the knock-on effects it would have on surveillance, privacy and young people’s wellbeing”, said OpenDemocracy.

Privacy and technology experts, as well as those working with children, have warned that the plans “could lead to a surveillance creep and data breaches”. They could also cut young people “off from social media’s potential benefits, such as giving LGBTQIA+ youth a chance to access communities”.

Social media companies have argued the ban could push children into unregulated parts of the internet and on to less safe sites and platforms. But Mark Dowey, whose son Murray died after being blackmailed on Instagram, told BBC Breakfast: “If that’s the best they’ve got then I think they’re in trouble. I think they’re basically acknowledging they don’t have a reasonable position here.”

Did it work in Australia?

The “key question” is whether it will actually work, said The Times. More than 70% of parents in Australia, which last year became the first country in the world to introduce a social media ban for under-16s, told the internet regulator their children were still on these platforms, a recent survey found. But supporters argue that the “problems there are about weak enforcement, not the model itself”.

Despite the decidedly mixed results of Australia’s prohibition experiment “the politics are broader: this is a culture-change moment, and a line in the sand from governments saying to tech companies: we make the rules”.

College graduates have been booing company bosses who mentioned artificial intelligence in graduation ceremonies as “AI anxiety” starts “boiling over into public backlash”, said Business Insider.

The trend is “highlighting a gulf” between older generations who feel the technology “offers new opportunities” and Gen Z, who are “growing increasingly anxious” about what it “means for their future”.

Backlash and resistance

A Gallup survey in April found excitement about AI among Gen Z has fallen from 36% last year to just 22%, while their anger towards the technology has risen by nine points, to 31%. Another survey, carried out by Numerator, found that among Gen Z people who don’t use AI, 57% are not open to adopting it, compared to just 32% of boomers.

“Read that again,” said Fortune – “older Americans are more open to AI than young ones”. It seems that a “surprising segment of the generation that was supposed to lead AI adoption” is actually “leading the resistance to it”. For them, AI was “foisted upon them” by their “parents, big tech CEOs” and Donald Trump.

“Every technology young people have ever loved”, like video games, social media and the internet itself, came to them as “play or transgression”, but AI “arrived as a mandate” from schools and employers. Also, Gen Z prizes “authenticity above almost everything” and AI “attacks” that.

Young people “were sold on the promise that a college education secured a good future”, but now employers are “gutting entry-level positions” in favour of AI, said Denison University student Jack Jackoboice in The Wall Street Journal.

A global survey found that 43% of CEOs plan to reduce junior roles, so young people are “actively being written out of a future” they have “no control over”.

A backlash is taking shape. Some Gen Z workers are “actively sabotaging their company’s AI initiatives” by feeding sensitive company data into public AI tools and by “intentionally producing low-quality, AI-assisted junk work” to make the technology “look unreliable”, said Vice.

Existential melodrama

But Gallup found that over half of 14- to 29-year-olds say they use AI daily or weekly, and some Gen Z-ers do see an upside in AI.

The “danger” is that “economic anxiety” can “curdle into existential melodrama”, said Ethan Tran, a student at Davidson College, in The Wall Street Journal. “Fear underrates human ingenuity”, so young people shouldn’t “hide from replacement” but “look for opportunities that arise from the transformation”.

The CEO of Big Machine Records, Scott Borchetta, also gave short shrift to AI anxiety, when graduates at Middle Tennessee State University booed him for saying that AI is rewriting the music industry. He told the hecklers: “You can hear me now, or you can pay me later.”

What happened

Apple on Monday unveiled an AI version of its Siri digital assistant at its annual Worldwide Developers Conference. The new Siri AI is the company’s response to OpenAI’s ChatGPT, Anthropic’s Claude and Google’s Gemini. OpenAI recently filed documents to prepare for a massive IPO, joining Anthropic and Elon Musk’s SpaceX-xAI.

Who said what

Apple is “betting the upgraded assistant can help close the gap” in the “crucial AI race,” but it has “taken a different approach from rivals,” Reuters said. Instead of pushing AI agents, the company “emphasizes practical features integrated into everyday tasks” and stressed that “personal data would remain private.” Analysts will be looking to see whether Apple’s “history of turning nascent technologies into popular products will apply to AI,” said CNN.

Some AI companies “appear to be racing forward, seemingly pursuing AI for the sake of AI, without clear regard for the ​people — all of us — that it’s ultimately meant to serve,” said Apple software chief Craig Federighi.

What next?

Apple is releasing its new “Golden Gate” software update — which includes Siri AI, more robust parental controls and other changes — immediately to developers, with a “public beta next month and a full launch to customers in the fall,” The Wall Street Journal said.

Artificial intelligence systems can now convince you they are human. Two large language models have passed the Turing test, which determines if a machine can “show the same intelligence as a human being,” said The Independent. This significant development in AI is troubling, as anthropomorphizing LLMs can lead to deception and raise questions about what’s real and what isn’t.

Man or machine

In the test, a person “engages in text-based conversations with both a human and a machine without knowing which is which," said Stanford University. If the individual cannot tell them apart, the machine is considered to have passed the test. Researchers tested four AI systems and found that newer LLMs can “effectively imitate people in short interactions,” said a study published in the journal PNAS.

“Given the right prompts, advanced LLMs can exhibit the same tone, directness, humor and fallibility as humans,” study author Cameron Jones said in a release. “While we know LLMs can easily produce knowledge on nearly every topic, this test showed that it can also convincingly display social behavioral traits, which has major implications for how we think of AI.” The four tested AI models were GPT-4.5 and Llama-3.1-405B, which were state-of-the-art models, as well as the older baseline models GPT-4o and ELIZA, a simple chatbot from the 1960s.

Of the models, “GPT-4.5 was judged to be the human 73% of the time, meaning interrogators selected it as ‘human’ significantly more often than they selected the real human participant,” said the release. Llama-3.1-405B, “given the same prompt, was judged human 56% of the time,” making it “statistically indistinguishable from the humans it was compared against.” The baseline systems performed significantly worse, with ELIZA being mistaken for human only 23% of the time and GPT-4o being mistaken 21% of the time.

No man’s land

AI models passing for humans is a concerning development. The Turing test is a “game about lying for the models,” Jones said in the release, and “one of the implications is that models seem to be really good at that.” A big risk of the existence of AI models with this ability is the rise of “counterfeit people.” Thanks to the ease of deception, we “need to be more alert,” and “people should be much less confident that they know they’re talking to a human rather than an LLM.” Still, AI is not yet at a level where it can be deceptive on its own.

While the bots did pass the Turing test, they also required specific instructions to do so. Each of the systems was “instructed to adopt a persona, or a specific character and communication style,” said The Independent. These prompts “worked partly by leading the systems to make mistakes in the same way a human would.” When the models were not prompted, they were much less likely to be mistaken for humans, and GPT-4.5 fell to a 36% win rate and Llama-3.1-405B to a 38% win rate. The models “have the ability to appear humanlike,” study co-author Ben Bergen said in the release, “but maybe not as much the ability to figure out what it would take to appear humanlike.”

With many Americans opposing the construction of giant AI data centers in their neighborhoods, some tech companies are proposing an unconventional solution: attaching mini data centers directly to people’s houses. At least one major startup backed by Nvidia is looking into the prospect, though it will likely be controversial.

How would these mini data centers work?

People typically associate data centers with big buildings churning out massive quantities of AI datasets. But the home version would be a “unit about the size of an air conditioner, mounted in the side yard,” said Scientific American. It could perform “artificial intelligence tasks, drawing power from your home’s energy supply” and theoretically “earning you discounted electricity and internet in exchange.”

Most of the attention has been focused on Span, an electrical panel startup that recently began manufacturing these types of units in partnership with Nvidia. The company said its mini data centers would be “less of a financial burden on residents” and “have a potentially lower ecological footprint than warehouse data centers,” said Fortune. Span’s units are also quiet, thereby “mitigating the problem of noise pollution that has drawn the ire of residents of areas with nearby warehouse data centers.”

Industry experts hope the home models like those proposed by Span could help alleviate the financial and energy constraints created by large buildings; a typical AI data center “consumes as much electricity as 100,000 households,” according to the International Energy Agency. Instead of “building a single large data center that requires its own substation upgrade or on-site gas turbines,” the AI “spreads compute across thousands of homes that are already connected to the grid,” said Scientific American.

What has the reaction been?

Creating more energy-friendly data centers is a “cool idea on paper, but it’s almost completely unproven in real-world use,” said Fast Company. And even if the home data centers took off, the “main point of resistance” is the fact that these centers “will result in higher electric bills for everyone in the area,” even if they are at people’s homes. Whether it’s a “new central data center or a distributed data center,” the “risk of higher costs — perhaps because of transformers and other infrastructure running hotter and degrading more quickly — could arguably be the same.”

Politically, gathering power from existing homes “may be easier than talking a city council into issuing a permit for a data center,” said Fast Company. But all of this is moot if tech companies are unable to perform the “tangled math of coordinating thousands of tiny residential energy resources to fuel the energy beast that data centers are,” said The Arizona Republic. While “distributed power generation has been around for years,” it has never “been harnessed at the scale needed for feeding data centers.”

Officials with Span remain optimistic that the home-based products will work. “There is certainly opportunity, as Span can provide homeowners with access to innovative technology and potential income generation that can help offset monthly energy costs,” a spokesperson for the company told Inc. “On a larger scale, if the technology proves out, it might also keep local infrastructure from being overburdened, which could keep land open for other uses, such as building homes.”

“Right now, it’s like the AI industry has a gas pedal but it doesn't have a brake pedal,” Anthropic co-founder Jack Clark told the BBC.

Anthropic recently overtook OpenAI, the makers of ChatGPT, as the world’s most valuable AI start-up. But Clark has called for a global freeze in AI development, warning that humans risk losing control of the technology. He revealed that 80% of the code that Claude, the company’s chatbot, is operating on was written by Claude itself. And reaching 100% is only a couple of years away.

Clark and his research colleague, Marina Favaro, have suggested that work at Anthropic could undergo “a meaningful slowdown or pause” if other AI tech firms were prepared to do the same. “If it were possible to effectively slow the development of this technology to give ourselves more time to deal with its immense implications, we think that would likely be a good thing,” they wrote in a blog post.

What did the commentators say?

Better regulation “would keep AI systems in their lane”, said David Krueger, a specialist in responsible AI, in The Guardian. We should insist companies have “clear and well-scoped purposes” for their AI tools, and “demand evidence that they are fit for purpose”. And they should report statistics and data so that we can see if their product is being used in ways that “deviate from its intended purpose”.

But the “safest, sanest” option is to “stop racing” to make AI smarter. The creation of Moltbook (a forum for AI agents that humans can only observe) is one of the “increasingly alarming warning signs” that “rogue AI agents” could be on their way. “We need to make sure” that rogue AI isn’t “capable of threatening humanity, by agreeing to enforceable, international limits on AI capabilities and AI development”.

There are some hopeful signs in the US. On Tuesday, Donald Trump signed a “much-awaited” executive order to establish a measure of vetting for AI companies, said Politico. It was “messy, muted and far less ambitious than Silicon Valley’s critics had hoped for” but it does mark a “sea change in Washington’s willingness to tighten” AI oversight. The new voluntary process of sharing new models with the US government, so that security risks can be identified and addressed before the technology is released, could “soon pave the way for mandatory vetting, federal pre-approval of advanced AI systems and other regulations”.

Some may think it “meaningful” that Trump is “doing something – anything – about AI”, said The Atlantic, but this executive order is “relatively toothless”. He wants to look like he’s being robust, to “score points” with the public, but, in fact “he is not saying or doing anything substantive at all”. The window for serious government regulation, anywhere in the world, is “rapidly closing”; “hopefully, it is not already gone”.

We’re missing the point, said John Burn-Murdoch in the Financial Times. “AI’s capacity to deliver genuine value has been vastly exaggerated.” In one US study, researchers tracking software developers before and after they adopted AI tools found an initial “explosive” increase in productivity (300% more files created or edited) but, after verification and review, just a 30% “uplift” in the number of releases. These are “powerful new tools” but it’s going to take some time before they can interact with current workflow “processes and structures” without friction or bottlenecks.

What next?

Trump’s executive order is a “good first move in creating a safer tech ecosystem”, said Jen Easterly, former director of the US Cybersecurity and Infrastructure Security Agency, in The New York Times. But a voluntary framework, predicated on mutual cooperation between private companies and the US government, “cannot guarantee” effectiveness. And, let’s not forget, a “principle enshrined in an executive order is only as durable as the administration that issued it”.

For this step to be a positive one, in an American context at least, the legislative branch needs to follow suit. The responsibility of building an AI environment that is “innovative, trusted and resilient” ultimately lies with the US Congress.

What happened

Florida on Monday sued OpenAI and CEO Sam Altman, alleging that the company’s AI chatbot violates state consumer protection laws. “Altman and ChatGPT have chosen the AI race over the safety and security of our kids,“ Florida Attorney General James Uthmeier (R) said at a news conference. “We’re going to make them pay for hurting our kids.”

Who said what

Uthmeier’s suit accuses OpenAI of a “litany of harms” driven by its “insatiable quest to win the AI arms race and amass large fortunes” regardless of known dangers, including abetting mass shooters, encouraging suicide and hooking minors on an unsafe tool. It’s the first state lawsuit against OpenAI and Altman, but the latest “broadside in a growing rebellion” against AI chatbots, The Wall Street Journal said.

Uthmeier has “emerged as a key antagonist” of AI since Florida’s GOP-led House “aligned with President Donald Trump” and blocked Gov. Ron DeSantis’ (R) “efforts to police” the technology, said Politico. Monday’s civil suit is separate from Ulthmeier’s ongoing criminal investigation into ChatGPT’s alleged role in planning a mass shooting at Florida State University.

What next?

Florida is seeking “more protections for children’s data and stronger parental controls” plus “financial penalties,” The Washington Post said. Uthmeier predicted other states will sue OpenAI as well.

Dating apps are “rooted in rejection and judgement” and that’s “not healthy”, Bumble founder Whitney Wolfe Herd told Fortune. She had an “epiphany” during a 14-month leave of absence that users are just “hurt people hurting people”, and has vowed to bring “more joy and satisfaction” to her app.

Bumble is now shifting to matching-making driven by AI – and it’s not the only dating app to see this as the solution to increasing dating-app fatigue. But newcomer Breeze is taking another route: switching the focus to in-person experiences by reducing opportunities to chat in app, and sending only a time-specific, limited number of matches.

‘Payment and consequences’

“Breeze is a welcome disruptor in the dating app landscape,” said Isabella Silvers in The Telegraph. Since it launched in Europe in 2020, after winning investment from the Dutch version of “Dragon’s Den”, it has clocked up more than two million downloads. Users join “matching pools” that bring together “like-minded daters”, based on everything from hobbies (“outdoor lovers”) to niche interests (“rat owners/lovers”). To date, the app has arranged more than 737,000 dates, “resulting in 10 babies – that it knows of”.

Users receive a “select number of profiles” at 7pm every day and the key to the app’s success seems to be “payment and consequences”. Once you accept a match, you have to fill out your availability and pay a £9.50 deposit to secure a drinks date (or £4.50 for a “walk and talk”), “before being allowed to make a decision on anyone else”. The chat function for matched users is only opened up four hours before the date – prompting last-minute date confirmations, rather than “meaningless messaging”.

Breeze is “evidently working”, especially in the Netherlands where it’s “the third most popular and fastest-growing” dating app, said Lydia Spencer-Elliott in The Independent. “But can it save Britain’s dismal dating scene?” It can certainly save us from “boring convos generated by ChatGPT”, or being stood up or ghosted or “strung out” for weeks with no follow-through. But “what it absolutely can’t save” us from “is ourselves”. It’s ultimately “knackering” to keep “riding the rejection rollercoaster of the dating-app landscape” – and, sometimes, “the best remedy is to give it all a rest”.

‘Charming chatbots’

There is “rampant” dating-app burnout, said Catherine Pearson in The New York Times. As Bumble embraces AI-powered algorithms to re-engage those who “crave an experience that feels less overwhelming and more purposeful”, it’s also removing its swipe feature. It’s hoping to “end superficial, snap judgements” by altering “the dating habits of millions of users who have grown used to vetting partners with the flick of a finger”.

But the AI pivot comes with risk. Integrating AI features “sloppily” could “alienate” dating-app customers, said Tatum Hunter in The Guardian. Some users are already reporting “being plagued by AI paranoia, unsure whether the people they are messaging are real or charming chatbots”. The messaging from the industry is clear: “if we let AI take the wheel, this will all get less depressing”. But can a “smooth, mindless path toward connection” really make dating more joyful?

Evolutionary psychology reminds us that “only a signal that is difficult to fake can carry reliable information about the sender”, said Andrew King on Quillette. A rightward swipe behind a screen “communicates almost nothing about the sincerity of the person making it”. But making an approach in person at a bar or an event carries the potential for “public rejection”, and that cost is a signal of sincerity. These signals “matter” and “cannot be easily digitised”: “the discomfort is the point”.

Getting you the right medical treatment more quickly – particularly if your life is at risk: that’s the aim of an NHS reform to unify patient records, so that doctors, nurses and paramedics can see a patient’s complete medical history, no matter where they are treated.

Single Patient Records could mean 20,000 fewer A&E visits and 6,000 fewer hospital admissions annually, said Health Secretary James Murray. This would save doctors about 500,000 hours, and the NHS £20 million, every year.

But plans for SPR, which come before Parliament today, face strong opposition from those who are concerned about the security of patient data and who will have access to it. We need to make sure that this pooled data cannot “be used inappropriately”, said the British Medical Association’s GP committee. “Ambitions to address fragmentation, improve productivity and reduce bureaucracy are laudable but they cannot come at the price of undermining confidentiality and public trust.”

What did the commentators say?

“The ambition is good,” said Alex Lawrence, a data specialist at The Health Foundation think tank. With its “Lego bricks” approach of stacking information together, SPR is the “most legislatively ambitious attempt” to “make care faster and safer” by getting patient data to “flow more freely” through the NHS system.

But “federating” the data and rolling out the system “is easier said than done”. It is still “unclear” what SPR will look like in practice, and “questions about how access, oversight and public choice will be managed remain unanswered”. Current data-sharing and confidentiality arrangements will be changed but key details – such as an individual’s right to restrict access to their records – have “been deferred to secondary legislation”. Its “absence on the face of the bill is a significant omission”.

“NHS digitisation projects have a chequered history,” said Laura Donnelly, health editor of The Telegraph. “A £12 billion programme for an NHS IT system in 2002 was abandoned” after 10 years, “due to spiralling costs and delays”. And Care.data, which was supposed to “extract GP records into a central database”, had to be scrapped in 2016 “following a public backlash over privacy concerns”.

Previous attempts to bring patient records together have been “beset by technical complexity, a mind-bending web of rules and roles, and some cultural intransigence”, said The Register. This time, the idea seems to be to use the current record systems in conjunction with the “controversial” Federated Data Platform run by US firm Palantir. “Either there’s going to be a new data store, which will be in Palantir, or there will be an infrastructure for bringing various independent APIs together” that uses Palantir’s FDP, Sam Smith, from data-safety campaign group medConfidential, told the news site.

There’s a reason why campaigners like medConfidential are calling SPR the “Single Palantir Record”, said investigative journalist Andrew Orlowski on Spiked. The company’s current contract with the NHS – which centres on using its FDP to improve efficiency – will be “worth over £1 billion if it runs its full course”. Palantir has had success in “winnowing” NHS waiting lists, but applying the singular goal of efficiency to patient data is “inimical to both interpersonal relationships – between patient and doctor – and trust”.

What next?

The plan is for SPR to be rolled out and made available on the NHS app as early as 2027. The Health Secretary has said that the Palantir contract was being reviewed ahead of its break point next year.

The NHS Modernisation Bill, which includes plans for SPR, as well as the abolition of NHS England, will have its second reading in the House of Commons today. Former Health Secretary Wes Streeting will speak from the backbenches to “back the bill he drafted”, said Donnelly in The Telegraph. He will no doubt “hail the changes” he made as health secretary and take “credit for the introduction of new AI tools and a funding uplift for GPs”. It’s a clear opportunity to boost his Labour leadership campaign.

What happened

A $9.7 billion Pentagon contract with Dell Technologies announced this week sent the company’s stock soaring, likely boosting President Donald Trump’s more than $1 million investment in the company, The New York Times and The Washington Post reported Thursday. “Government ethics watchdogs are sounding the alarm” not only because Trump “potentially stands to gain financially” from the Dell deal, the Post said, but also because he “has repeatedly praised the company at public events” since acquiring the shares earlier this year.

Who said what

The Dell investments were among more than 3,600 trades executed in Trump’s investment portfolio from January through March, according to a mandatory filing released this month. The Trump family has “argued that the president does not personally control the trading,” but the president’s financial accounts “are not in a traditional ‘blind trust,’” the Times said. And his Dell purchase “draws new attention to the inherent problems” with the family’s “widespread investments” in military drones, cryptocurrency, mining and prediction markets while Trump “oversees policy and government purchase decisions for those same sectors.”

What next?

Presidents are exempt from an ethics law that prohibits official self-enrichment. Congress should “revisit the arrangement whereby we rely on the president’s own sense of integrity rather than law to avoid conflicts of interest,” Greg Williams from the Project on Government Oversight told the Post.

The rise of artificial intelligence depends on the construction of giant new data centers to supply the necessary computing power. But Americans do not want the facilities in their neighborhoods.

Backlash to data centers is “bipartisan and growing across the country,” said 404 Media. States and cities are outlawing the “noisy, power and water hungry buildings” in a fight that could “shape American politics for years to come.” Seven in 10 Americans oppose building a data center in their area, said Gallup, higher than the 53% who would oppose a nuclear plant nearby. Industry leaders are now fretting over their inability to win public opinion that is “increasingly aware and skeptical,” said Business Insider. The tech sector “hasn't done a good job of explaining itself,” said Flexential CEO Ryan Mallory, whose company develops and operates the data centers.

What did the commentators say?

The backlash to AI “could get very ugly,” Lila Shroff said at The Atlantic. A “record number of proposed projects” were canceled during the first quarter of this year after “local pushback.” In April, an Indianapolis councilman found a “NO DATA CENTERS” note under his doormat after somebody shot at his house 13 times.

The fights over data centers will likely only “intensify,” as the facilities “stimulate local economies” but also take “physical and environmental tolls” on the places they are built, said Shroff. And though AI opponents may not be able to stop Anthropic from distributing its Claude model, “they can raise concerns about new construction at a local city-council meeting.”

“Nobody wants this in their backyard,” Sara Pequeño said at USA Today. In Utah, officials overrode local opposition to approve a giant new center that will consume “more than two times the energy used in the entire state.” Rural areas across the country face similar proposals. Data centers are “almost certainly here to stay” because of the computing power needed to keep up with “our ever-growing reliance on AI.” But Americans “clearly don’t feel great” about having them nearby.

The “brewing populist resistance” to data centers is a “critical new front in the fight against tech-enabled authoritarianism,” Astra Taylor and Saul Levin said at The Guardian. A local fight over land use can double as opposition to “job-eating algorithms, distorting deep fakes and autonomous drone strikes.” It also portends the next big electoral fight. AI is “shaping up to be a key fault line” in both this year’s midterms and in 2028.

What next?

The canceled data center projects are “sapping confidence” among AI investors, the investment bank Jefferies said in note to clients, per Axios. The pushback could become a “financial liability for AI labs if it continues to curb access” to the computing power artificial intelligence requires, the outlet said.

The backlash movement has one notable new ally. Erin Brockovich, the activist portrayed in an Oscar-winning performance by Julia Roberts, has launched a new website tracking proposed and under-construction data centers. The map “captures the real-world footprint” of the AI race, she said on the site.

What happened

Pope Leo XIV on Monday released his first encyclical, “Magnifica Humanitas” (“Magnificent Humanity”), making a practical and moral case for “safeguarding the human person in the time of artificial intelligence.” AI “needs to be disarmed” as “an instrument of domination, exclusion and death,” the pope told a packed hall at the Vatican. “It must be at the service of all, and of the common good.”

Who said what

Addressed to “all people of good will,” Leo’s “methodical” teaching document traced the Catholic Church’s established “social teaching and applied its core concepts,” including solidarity and the dignity of work, “to the digital revolution,” The Associated Press said. The document’s title “says it all,” The New York Times said: Leo is “less interested in technology than in humanity.”

“Technology should not be considered, in itself, as a force antagonistic to humanity,” Leo wrote, but “the pursuit of greater profits cannot justify choices that systematically sacrifice jobs.” AI’s growth needs to be guided by “robust legal frameworks, independent oversight, informed users and a political system that does not abdicate its responsibility,” he said. “A more moral AI is not enough if that morality is determined by a few.”

What next?

Tech and religion experts said Pope Leo’s encyclical “will likely become a benchmark in the debate over AI, a point of reference for policymakers, researchers and ordinary folk alike,” the AP said. The pope is “really doing the Lord’s work here, and I say that as an atheist,” humanist Harvard chaplain Greg Epstein told the Times.

A hypothetical doomsday for quantum computing could be on the horizon, computer scientists have warned for decades. But cybersecurity experts are now racing against the clock after Google announced that this “Q-Day” could be here much sooner than originally anticipated.

What is ‘Q-Day’?

It is the hypothetical day that quantum computers will acquire “enough resources and stability to crack conventional cryptography,” said CNN. When that day arrives, it could spell disaster for millions of people’s private information, as “every financial transaction, medical file, email, location history and crypto wallet protected by today’s commonly used algorithms could be unlocked.”

Unlike conventional computers, quantum computers utilize “quantum-mechanical phenomena” that allow them to “perform calculations that are practically impossible for even the most powerful supercomputers today,” said Forbes. Experts believe these computers could eventually crack RSA cryptography, the algorithm of prime numbers that helps to safeguard encryption. Some fear this could be accomplished “not in billions of years but in hours or days.” Others believe some “bad actors may already be collecting encrypted data” in secret, said CNN.

It was previously believed that Q-Day was still far into the future, giving the tech world plenty of time to prepare new safeguards. But Google recently announced it believes the day could arrive as soon as 2029, and the “new estimate means that governments, companies and other entities may have far less time to prepare,“ said CNN. Many are comparing Q-Day with “Y2K, or the millennium bug, a computer flaw that programmers thought might cause severe systemic problems after Dec. 31, 1999.”

What can be done?

Many companies are being urged to boost their cybersecurity initiatives as the potential for Q-Day looms. Google is also creating guidelines it hopes will “provide the clarity and urgency needed to accelerate digital transitions not only for Google but also across the industry,” the company said. To accomplish this, Google “specifically is pushing for a transition to post-quantum cryptography, or the use of new, quantum-resistant algorithms to secure data against future attacks,” said Barron’s.

Even if the 2029 date doesn’t come to pass, there is still a 10% chance Q-Day will occur by 2032, Justin Drake, a bitcoin security researcher who published a paper on the matter, said on social media. No matter the date, other precautions are being taken. For example, cryptographers “have devised new encryption algorithms that rely on problems that quantum computers don’t have an advantage over classical computers in solving,” said Ars Technica. The National Institute of Standards and Technology has also “advanced several algorithms that have yet to be broken and are presumed to be secure.”

Government entities have been weighing in too. In 2022, the National Security Agency (NSA) announced a plan to boost Q-Day readiness by the 2030s. But recently, the deadline “has been in flux as both the Biden and Trump administrations have issued executive orders prioritizing quantum readiness,” said Ars Technica. The NSA is currently “adhering to a 2031 deadline.” Despite these plans, experts remain worried, as encryption is “not a permanent state of protection,” said Forbes. It is a “time-locked safe that someone may already be holding, waiting for the combination.​”

Google has become so synonymous with online search that its name has evolved into a verb in its own right. Now, the company is attempting to “revamp its decades-old business model to fit the era of artificial intelligence”, said CNN. In essence, “Google wants to help you google less”.

‘New era’ for search

Although Google already offers “AI Mode”, it will now integrate the technology across the entire search experience through its new Gemini 3.5 Flash model. Rather than simply typing keywords or short phrases, users will be able to ask conversational questions, share images or voice commands with agentic AI, and even interact through live video.

Instead of generating only the familiar list of blue links, Google Search will give a customised AI-written summary of the topic being researched. This will then open a conversation with AI Mode directly on the main search page, allowing users to ask follow-up questions more naturally.

This marks a “new era for AI search”, according to a Google blog post, bringing “advanced model capabilities” and “new AI features” to Search. The update will allow users to deploy AI agents “just by asking a question”. The company is also introducing a new intelligent, AI-powered search box, which it describes as Google’s “biggest upgrade in over 25 years”.

Crucially, the shift means that search will become more conversational and personalised, reducing the need to click through to web pages. Increasingly, Google will function more like an assistant than a traditional index of third-party information providers.

‘Radical transformation’

For many people, Google’s search box is the “lobby of the internet”, said Time, so this “radical transformation” signals a major shift in how people use the web. It could “disrupt many industries” that rely on search traffic to attract customers, with news publishers and small businesses particularly vulnerable.

The changes will likely “further decimate” referrals from Google to publishers, which have “already been suffering from declining referrals” because of AI Overviews, said TechCrunch. The trend has already “put some ad-dependent media operations out of business, and now things will likely get worse”.

Using AI-based searching could also erode important skills, said Riley MacLeod on internet news site Aftermath. Google Search is “one of the first and primary places that people experiment with and grow their information-searching skills”. While “spoon-feeding” users AI summaries and “obscuring or bypassing the source of the information” may seem convenient, it risks depriving people of the opportunity to build the “vital information literacy skills” they “need more than ever in an AI-obsessed world”.

For Google, however, the ambition is far larger: to move “closer” to its long-term goal of developing artificial general intelligence – a “theoretical stage of AI” where technology becomes as intelligent as humans across a broad range of subjects, said CNN. The competition is intense, with OpenAI, Meta and others all “racing to be the first to get there”.

The sound of typing has been the background hum of office work for a century and half. But now it’s all about whispers.

After years of bashing typewriters, then tapping keyboards, desk-bound employees are, in ever-increasing numbers, murmuring to AI dictation apps to send emails, draft reports and write code.

‘Double words per minute’

Voice-to-text software has been around since the 1960s but it was always “clunky” and slow and “never worked quite right”, said employment reporter Jo Constantz on Bloomberg. But now advances in AI have made it “viable”: it can “take the messiness of speech and package it into something more useful”.

Early adopters of AI dictation apps are “drawn inexorably to the promise of ever-greater productivity”. In “voice mode”, you can produce double the words-per-minute than you can when typing.

Dictation is definitely “having a moment”, said Joe Castaldo, business reporter at Canada’s The Globe and Mail. More and more software engineers, in particular, are switching from “pressing keys individually” to “adopting AI-powered speech-to-text apps to verbally issue instructions” to tools such as Anthropic’s Claude Code. Eight months ago, internet entrepreneur Reid Hoffman posted on his LinkedIn platform that he has been “voicepilled”: he’d realised you can “amplify your ability” by “seriously using your voice to interact with technology”.

Start-ups today are like “a high-end call centre – except everyone is chatting with AI”, one venture capitalist told The Wall Street Journal. There is an “etiquette”: “users try to keep their voices low and often wear headphones to block out sound from their dictating neighbours, dialling down the annoyance factor”. But talking to yourself is still “weird, if not a little embarrassing”.

‘Velocity towards voice’

It’s too early to say if and when “the Qwerty keyboard might follow the ticker tape and fax machines into obsolescence” but “the velocity towards voice is accelerating”, Dylan Fox, CEO of San Francisco-based AssemblyAI, told the Los Angeles Times. We’re predicting a 10 to 100-fold “increase in demand for voice, AI applications and interfaces”.

There’s now “a mad dash to dominate any corner of the evolving field”, said Bloomberg’s Constantz. The market for AI voice generators alone is estimated to be worth $7.7 billion (£5.75 billion) this year, rising to $21.8 billion (£16.27 billion) by the end of the decade, according to US consulting firm Grand View Research.

Google, Apple and Microsoft have invested heavily in their voice-to-text products, and dictation app start-ups – many with variations of “whisper” in their name – have experienced remarkable growth over the past year. After all, Superwhisper founder Neil Chudleigh told The Globe and Mail, “we’re talking about replacing every keyboard on the planet”.

Restrictions on high-risk uses of artificial intelligence in the EU will be delayed by more than a year under a deal struck by its legislators.

The deal “marks a notable rollback” in the bloc’s “digital rulebook after years of Brussels proudly marketing itself as the world’s tech cop”, said The Register.

What is changing?

The EU’s AI Act came into force in August 2024 after “years of talks”. But as part of a “phased rollout”, the rules governing high-risk uses were only “set to kick in this August”, said Politico.

Instead, the bloc has “hit the regulatory equivalent of ‘snooze for 16 months’”, said The Register. “The headline change pushes back enforcement of rules covering systems” in areas such as biometrics, critical infrastructure, education, employment, migration, and border control until December 2027.

For products like lifts and toys, compliance deadlines for their AI systems are “stretching” further – to August 2028. Meanwhile, smaller companies get “more breathing room”. The EU hopes it will “avoid duplication between sectoral and AI rules”, it said in a press release.

EU officials insist the delay is “about timing, not watering down the law”. They claim the rules are “moving faster than the standards needed to support them” and that companies currently “lack the guidance and technical tools required for compliance”.

Is this a win for Big Tech?

The change of heart is a “big win” for tech firms and industry groups that have been lobbying the EU to “soften” the AI Act, said The Register. As recently as last week, bosses from companies including ASML, Airbus, Ericsson, Nokia, SAP, Siemens and Mistral AI “publicly warned that Europe risked over-regulating itself out of the global AI race”.

The new deal, which marks the “first significant rollback” of rules in the digital sphere, came after the EU faced pressure from the US over its tech laws. There were also “warnings” from its own industry and governments that “strict restrictions had put the bloc at a disadvantage in a global AI race”, said Politico.

“Only a couple of countries around the world” followed the EU’s lead on restrictions, so the bloc “faced criticism” for “cracking down on AI too early”, despite “civil society” saying that “rules are needed to protect people from the potential harms of the emerging technology”.

Arba Kokalari, a Swedish MEP on the internal market committee, insisted that the EU is “not weakening any safety rules”, but rather “clarifying the rules for companies in Europe”.

What is staying the same?

Some aspects of the AI Act will keep to their original schedule. Bans on unacceptable-risk AI have applied since February 2025, according to the European Commission. The transparency obligations under Article 50, including disclosure for chatbot interactions, will come into force from 2 August.

The European Parliament and Council also agreed to ban AI systems that create child sexual abuse material or that depict identifiable people in sexually explicit content without consent. Companies have until the end of this year to comply.

Palantir Technologies Inc., a Miami-based company that specialises in data integration and analysis, is seldom out of the news. This is partly because it works in controversial sectors: its biggest client is the US military, and its software is used in conflicts from Israel to Ukraine. Clients also include the CIA and US Immigration and Customs Enforcement (Ice); it was involved in Elon Musk’s short-lived Department of Government Efficiency.

It has also expanded into healthcare: in Britain, its contracts include a £330 million deal with NHS England, as well as a £240.6 million deal with the Ministry of Defence.

But its notoriety is in part because of its eccentric CEO, Alex Karp. Palantir recently posted on X/ Twitter a manifesto penned by Karp, which, among other things, declared that “Silicon Valley has an affirmative obligation to participate in the defence of the nation”; called for universal conscription; encouraged the development of AI weapons; and condemned the West’s “vacant and hollow pluralism”. One MP called it “the ramblings of a supervillain”.

Where did Palantir come from?

Founded in 2003 by a group of tech moguls headed by Peter Thiel, a co-founder of PayPal and a libertarian political activist, Palantir was named after the “seeing stones” in “The Lord of the Rings”. (Thiel is a J.R.R. Tolkien fan.) Originally, it applied PayPal’s fraud detection system – which successfully identified fraudulent activity on eBay – to US national security; early funding came from In-Q-Tel, a venture capital firm that funds projects for the CIA.

Palantir’s technology was taken up by the US defence establishment under President Obama – it is rumoured that it was involved in the assassination of Osama Bin Laden – and it helped the US and UK governments with contact tracing and vaccine distribution during the Covid pandemic. It now helps the Trump administration track undocumented immigrants, and provides Israel’s military with “intelligence and surveillance services”. Palantir currently has a market capitalisation of some $350 billion.

What does it actually do?

One former employee likened Palantir’s work to “really extravagant plumbing with data”. Most big companies and government agencies have a lot of information they can’t easily use because it’s stored in a hodgepodge of different systems and databases.

Palantir’s core products – “Foundry”, primarily for civilian use, and “Gotham”, for military and law enforcement – sit on top of those different systems and pull all the data together in an interface that’s easy to use (little coding is required). A big selling point is that Palantir doesn’t itself access or exploit the data, which stays with the customer; it just makes it easier to analyse. This is useful for all sorts of unobjectionable things, such as Covid testing and tracing. But it also allows Ice to collect large amounts of information to investigate individuals – and it helps the US military to plan bombing campaigns.

What is its military role?

Palantir is the leading contractor for Project Maven, the US military’s (and Nato’s) targeting system. Maven draws together a mass of data from drones, satellites, signals and other sources to flag potential targets; it presents findings to human analysts in one clear user interface; and can relay their decisions to appropriate weapons systems.

According to a new book, “Project Maven” by Katrina Manson, the entire “kill chain”, from target identification to target destruction, consists of four clicks. Maven allows hundreds of targets to be hit per day; and adding in AI tools to help interpret data means that number is capable of rising into the thousands.

Similar Palantir technology is used in Ukraine, and since 7 October 2023, it has worked closely with the Israel Defence Forces, whose AI-assisted systems use algorithms to identify and assassinate suspected Hamas agents.

What are the implications of this technology?

Speeding up the steps between identifying a target and destroying it is fundamental to modern warfare, so it is immensely valuable. In Ukraine, Palantir’s tools have helped to fuse battlefield intelligence, track and destroy drones, even document war crimes.

But such systems are not infallible, and accelerating the kill chain also minimises the role of human judgement: Maven was used to wrongly identify a primary school in Minab, Iran (in a building used years before by the Revolutionary Guard Corps), as a military target. US missiles killed some 168 people, mostly young girls.

Where does the NHS come into all this?

Palantir has been involved in the NHS’s data-handling since 2020, during Covid. In 2023, it won a contract to develop the Federated Data Platform, designed to streamline tangled datasets across the NHS and help clear hospital backlogs. In some hospitals, for example, scheduling operations may require staff to consult separate systems for waiting lists, theatre bookings, staff rotas and equipment orders.

But many critics dislike the idea of a US spy-tech firm, with links to the US and Israeli militaries, potentially gaining access to sensitive health data. Others question its value for money.

How worried should we be?

Palantir has become “a cultural shorthand for dystopian surveillance”, says Wired magazine. It is a cause célèbre on the British Left that has been taken up by the Greens’ Zack Polanski. Arguably, though, it is just a data analytics company with a militarised culture designed in part to give it a mystique: the company’s slogan is “We build software that dominates”; it uses military and intelligence jargon instead of more standard office terms. (Its data consultants are known as “forward deployment software engineers” or “deltas”.)

But not least because of its close links to a US administration that is an unreliable ally at best, many policymakers in Western Europe are now reconsidering the wisdom of using Palantir’s services.

Who is Alex Karp?

Karp, 58, the son of a Jewish doctor and an African-American artist from Philadelphia, was a left-wing student activist; he studied in Frankfurt under the socialist philosopher Jürgen Habermas and has no background in computing. He had become friends with Peter Thiel at Stanford Law School, and in 2003 helped co-found Palantir.

Karp has always been outspoken about the company’s values – Palantir has long refused to work with Chinese or Russian companies – but these have moved markedly to the right over the years, and today he often rails against “woke” thinking, describing it as “pagan”. Karp is a fan of martial arts and pistol shooting, and has a retinue of bodyguards drawn from Norwegian special forces, apparently because they are able to keep up with his obsessive cross-country skiing. His net worth is estimated at over $15 billion.

Palantir’s “manifesto”, like Karp’s recent book “The Technological Republic”, seemed to argue for a merger between Silicon Valley and a nationalistic, militarised US state; but it also railed, idiosyncratically, against the iPhone and the “post-war neutering of Germany and Japan”. It was seen by some as an attempt to curry favour with the Trump White House, which has turned on tech firms deemed unsupportive, such as Anthropic.

It might be the ultimate clash of tech giants. Elon Musk and Sam Altman are in court this week, battling over the origins of OpenAI and its pivot from a nonprofit organization to a for-profit business. It’s a “deeply personal” civil trial, said The New York Times, featuring “two very different tales” of OpenAI’s founding.

Musk helped start the company as a nonprofit and contends it was “ripped from its promise of altruism” by Altman’s greed. It’s “not OK to steal a charity,” Musk said on the witness stand. Altman, OpenAI’s CEO, counters that the lawsuit is simply “sour grapes” for the success of OpenAI’s ChatGPT years after Musk parted ways in 2018, said the Times. Altman and OpenAI “had the nerve to go on and succeed without” Musk, said William Savitt, OpenAI’s lead counsel.

What did the commentators say?

The trial is “big in every conceivable measure,” said Slate. Musk is seeking $130 billion in damages along with the removal of Altman and another OpenAI co-founder, Greg Brockman, from the company’s board of directors. It also comes as both OpenAI and Musk’s SpaceX — which houses his current AI venture, xAI — prepare to take go public. The verdict “could change the very future of Silicon Valley and the future of tech throughout the world forever.”

Altman and Musk “sure dislike each other,” Matteo Wong said at The Atlantic. Altman and Musk founded OpenAI because they disagreed with Google’s approach to artificial intelligence then split up over their own disagreements. The trial is giving the public its “clearest glimpse” at a small clique of tech pioneers “whose bickering is shaping the most expensive infrastructure buildout in human history.” It is a technology that could “upend the labor market” and “reshape the geopolitical order,” and neither man wants the other to have that kind of power. The trial makes the AI boom “seem sordid and small.”

A “yearslong feud” between Altman and Musk means the trial is “going to get messy,” Elizabeth Lopatto and Hayden Field said at The Verge. Musk appears to be “trying to damage OpenAI’s reputation however he can.” His demands that the company change its operating structure and remove executives “are likely unrealistic.” But if enough ugly secrets are revealed at trial, Musk will “have made it look like it’s not worth keeping Mr. Altman in his position” at the top of OpenAI, Georgia Institute of Technology’s Deven Desai said to the outlet.

What next?

The trial comes at a “precarious moment” for OpenAI, Rob Nicholls said at The Conversation. Altman was recently the subject of an embarrassing profile in The New Yorker, and the company is “bleeding” money as rival Anthropic surges to the front of the AI conversation. OpenAI expects to lose $14 billion in 2026 and recently shut down its Sora video-creation product. A Musk victory might derail OpenAI’s IPO and leave “ripple effects” that “could be felt for many years to come.”

Apple founder Steve Jobs created the iPhone and cultivated a rockstar reputation for innovation. His successor, Tim Cook, turned the company into a globe-spanning colossus of profit. What will the next CEO, John Ternus, do to build on their legacies?

The 51-year-old Ternus “knows Apple at its core” after a quarter-century at the company, said CNN. As a senior executive since 2021, Ternus “led the hardware engineering behind Apple’s most recognizable products” like the iPhone and iPad and was “essential” in developing the new mid-price MacBook Neo.

His promotion to CEO “isn’t much of a surprise,” given that he had been seen as a front-runner to succeed Cook “for at least the last year,” CNN said. His task is to position the company for further success in the age of artificial intelligence. Ternus faces pressure to “produce success out of the gates,” Wedbush analyst Dan Ives said in a note.

An ‘Apple lifer’

Ternus is a “safe choice in a dangerous moment” for Apple, said Semafor. Cook replaced Jobs when Apple was at the “height of its influence” and built it into the first company with a $1 million market cap. The company is “still a financial juggernaut” though it does not command its former cultural cachet. Ternus is an “Apple lifer” unlikely to take Apple in a “radical new direction” that would “squander its lucrative business.” But his ascension comes as AI transforms the “entire concept of computing and technology.”

The “defining challenge” for Ternus is “fixing the company’s AI strategy,” said CNBC. Apple has so far avoided “hefty capital expenditures” on AI data centers and “punted” on its own AI model. Instead, Apple has bet that consumers will use its iPhones and other products to run AI. Choosing Ternus as CEO signals the company’s belief that the “future of AI will run through tightly integrated devices, not just software,” the University of Notre Dame’s Timothy Hubbard said to CNBC.

Apple faces an “existential challenge” figuring out “what comes after the iPhone,” said Axios. Cook “executed masterfully” to maximize iPhone’s success but “largely sputtered” with new products like the Vision Pro and a failed attempt at building autonomous cars. Companies like Meta and Google are pushing smart glasses, and former Apple design guru Jony Ive is designing hardware for OpenAI. A new leadership era opens with Apple “chasing its next hit” product. Cook demonstrated that Apple can grow. Ternus instead “must prove that it can still innovate.”

Making ‘first-rate physical things’

Apple has put the “hardware guy in charge,” said The Wall Street Journal, and is betting on itself as a “maker of first-rate physical things” in an AI-dominated world. That means navigating “complex geopolitics threatening Apple’s supply chain” and countless “regulatory battles around the world.”

Ternus is expected to “bring back Jobs-era decisiveness” to Apple’s CEO suite, said Bloomberg. Cook was known for “incrementalism” in moving the company’s product line forward, Forrester Research’s Dipanjan Chatterjee said in a note. Ternus “must define Apple’s future as ferociously as he defends its past.”

What happened

Meta said Thursday it will cut about 8,000 jobs, or 10% of its workforce, as it shifts resources to artificial intelligence. In a company memo, Chief People Officer Janelle Gale said the social media behemoth would also close 6,000 open positions “as part of our continued effort to run the company more efficiently” and “offset the other investments we’re making.”

Who said what

CEO Mark Zuckerberg is “reorganizing his company around AI products in a fierce race” against OpenAI, Anthropic and Google, The New York Times said. Zuckerberg has “made no secret of his AI ambitions,” including rolling out AI-powered social media he “hopes people will incorporate into their daily lives,” and he has pushed employees to “use AI in their daily work.”

Meta’s cuts are the “latest in a string of tech industry layoffs fueled” by AI’s efficiency promises, CNN said. Amazon said it would cut 16,000 workers in January, and financial-tech firm Block’s 40% workforce cut in February “came with a stark warning that more companies would follow suit.” Microsoft on Thursday said it was offering buyouts to 7% of its workforce to invest in AI.

What next?

Meta said it will notify employees being laid off on May 20.

Allbirds is making a complete heel turn after the shoe brand announced its pivot to AI. And many are skeptical that the footwear company will succeed in making such a big switch to the convoluted tech space. But Allbirds is just the latest in a list of companies that got their start in one industry, then changed to something quite different.

Android

Android cellphones have become as ubiquitous as iPhones in modern years, but the company didn’t start out in the phone game. The brand was launched in 2003, originally “conceived as an operating system for digital cameras,” said software development company Velvetech. By the time Android got up and running, the “market for digital cameras significantly fell,” whereas the “mobile device market was constantly growing.”

The company was forced to pivot to stay alive and began producing an operating system with more widespread uses. It is now used “primarily for mobile devices such as smartphones, tablets, smartwatches and other wearable devices,” said IT brand Spiceworks.

Nintendo

Nintendo has always made games but probably not the kind you’re thinking of. The company was started in 1889 when its founder, Fusajiro Yamauchi, began producing Japanese playing cards called Hanafuda in Kyoto. By 1902, Yamauchi “started manufacturing the first Western-style playing cards in Japan,” said Nintendo’s website. The company began growing in size throughout the mid-20th century.

By the 1970s, Nintendo realized it had to make a change to keep up with the times, and in 1975 “began the development of its first electronic video game systems,” said BBC News. In 1978, Nintendo “produced a computer game version of the board game Othello” and has since been responsible for producing some of the most iconic video games franchises of all time, including Mario, The Legend of Zelda and Donkey Kong.

Nokia

Nokia has made perhaps the biggest one-eighty of the companies on this list. While known today for its industrial-strength cellphones, the company started in the 1860s as something wholly different: a wood pulp mill in Finland. This mill was the first step in the mass production of paper. The modern company was eventually formed as a “merger between the Nokia Company (paper), Finnish Rubber Works and Finnish Cable Works in 1867,” said the Crypto Museum, a Dutch virtual museum.

Prior to its eventual focus on cellphones, Nokia became a bit of an everything brand. It has been “involved in the production of paper, rubber, electricity, car and bicycle tires, footwear, communication cables, television sets, consumer electronics, personal computers, robotics, capacitors, plastics, aluminium, chemicals, mobile phones and last but not least: military communications equipment,” said the Crypto Museum.

Slack

Slack is used today as a business-to-business chat tool by numerous companies and industries. Yet it originated in the 2010s as an “internal communication tool” for the “quirky online multiplayer game Glitch,” said Encyclopedia Britannica. The video game garnered positive reviews, but its “creators found the game to be expensive and unwieldy.” They soon started looking for alternative ways to implement the technology.

This arrived in the form of a rebrand: Slack, a “provider of a messaging tool for facilitating workplace communication, an ‘email killer’ and the ultimate collaboration app,” said TechCrunch. Today, Slack is “used by more than 100,000 organizations, including 77 of the Fortune 100 companies, demonstrating the network effect of a mature and innovative product,” according to the company itself.

Volkswagen

Volkswagen has always sold cars. But in this case, it’s the company’s history that represents a major redirect. The brand is well-known for its associations with the Nazis during World War II: In 1937, Adolf Hitler’s party “founded a state-owned company that was later named Volkswagen, or ‘The People's Car Company,’” said NPR. Volkswagen leadership would eventually disavow its Nazi ties.

The pivot came in modern times, as Volkswagen shifted from supporting antisemitic Nazi Germany to negotiating weapons deals with the state of Israel. In a tinge of irony, Volkswagen, which “produced parts using forced labor for V-1 cruise missiles used by the Wehrmacht during World War II, may soon be manufacturing parts for an Israeli-designed missile defense system,” said Haaretz.

YouTube

YouTube is best known as the video platform where you can watch just about any kind of video. But it was originally started in 2004 by three PayPal employees who had an “idea for a website for users to upload video dating profiles,” said Business Insider. The company was even trademarked on Valentine’s Day. As a dating site, YouTube “attracted little interest, forcing the co-founder to take out ads paying women $20 to upload dating videos.”

Then people began “uploading videos of all kinds to YouTube,” said Business Insider, and the website took off as a general platform. Today, over “20 million videos are uploaded daily” on YouTube, with an estimated 20 billion total videos on the site, the company said.

Claims that new AI models can outperform humans at some hacking tasks has sparked widespread alarm about the future of digital security.

Tech firms “usually create buzz around products they plan to release”, said The Economist. American artificial intelligence lab Anthropic, “has managed to create excitement – and a good deal of worry – around something it plans not to”, having announced that its new Claude Mythos model would not be released to the general public.

The problem is not that the new model is “buggy or unreliable” but rather “that it works so well that releasing it would put the world’s digital infrastructure at risk”.

What did the commentators say?

This next generation of AI models such as Anthropic’s Mythos or OpenAI’s new closed-version GPT 5.4-Cyber can not only write code, but also recognise errors – or “bugs” – in the code, which can be used to both identify potential weaknesses but also ways to attack computer systems.

“It’s impressive – and, at the same time, worrying” – because it makes cyberattacks “easier”, said professor of cyber security Florian Tramèr on ETH Zurich university’s website. A lone hacker “can suddenly try out thousands of variants” and “if one attack fails, he or she can simply try with the next one.” “This increases the risks for companies, state institutions or even private individuals,” especially “if such models become cheaper and more efficient”.

Recognising the danger this might pose, Anthropic has limited access to Mythos to a handful of trusted tech companies under an initiative called Project Glasswing. Similarly, OpenAI is providing limited access to GPT-5.4-Cyber to vetted security professionals so they can use it for defensive cybersecurity measures.

Yet even Anthropic’s strict security protocols appear to have been breached, after the company confirmed it was investigating how a group of users gained “unauthorised access” to Mythos Preview “through one of our third-party vendor environments”.

The risk of unauthorised access will only “add to anxiety” about Mythos, and “raises concerns” about whether Anthropic “can keep the technology it develops out of the hands of bad actors”, said Cristina Criddle in the Financial Times.

News of these new models’ cybercapabilities had already “sent shockwaves through the markets and prompted high-level discussions among financial institutions and global regulators”, with finance ministers from across the G7 hosting bank bosses to discuss what AI-enabled hacking might mean for their businesses.

What next?

Capitalising on a “mix of fear and excitement over AI and its future impact” has “become a hallmark of the sector and its marketing strategies in recent years”, said BBC reporters Liv McMahon and Joe Tidy.

In the case of Mythos, “we still do not know enough about it to know whether these hopes or fears are justified, or more a reflection of the hype surrounding the industry”.

In reality, “like other tools from the long history of cybersecurity”, the latest AI models “can be used for both offence and defence”, said The New York Times.

There is still disagreement on “whether one side of this struggle has gained a significant advantage through AI” and experts are “unsure how the battle will play out in the coming years”. Most agree, however, that “the companies and governments that do not embrace the latest AI for defensive purposes will leave themselves enormously vulnerable”.

With the cyberenvironment experiencing the “most change” ever, said Francis deSouza, the chief operating officer and president of security products at Google Cloud, “you have to fight AI with AI.”

Amazon founder Jeff Bezos is “close” to securing a $10 billion (£7.4 billion) fundraising deal from investors for his AI lab, codenamed Project Prometheus, said the Financial Times. The deal would make the company, which aims to explore how AI systems can be applied across physical industries, “one of the best-financed early-stage start-ups globally”, and marks the first time Bezos has served in an operational role since stepping down as chief executive of Amazon in 2021.

Project Prometheus will propel Bezos into the ranks of the AI titans heading firms with multi-billion-dollar valuations, such as Anthropic, OpenAI and Palantir. With the industry elite divided by ongoing legal feuds and conflicting political ideologies, the personalities of the individual CEOs look set to shape the course of AI as much as the technology itself. Here are the five names to watch.

Sam Altman

The OpenAI CEO is more and more becoming the “protagonist” of our times, said Lily Isaacs in The Observer. As with Faust, Victor Frankenstein and J. Robert Oppenheimer, we are beginning to “share the uneasy feeling that enlightenment carries within it the seeds of catastrophe”.

Launched by OpenAI in November 2022, ChatGPT is the chatbot that has “redefined the standards of artificial intelligence”, said Forbes. As the company nears a possible value of more than $1 trillion (£740 billion), “one of the biggest so-called risk factors” to the company is “Altman himself”, said Dave Lee in Bloomberg. Altman was fired by the board in November 2023, only to be reinstated days later.

Reading the year-and-a-half-long investigation by Ronan Farrow and Andrew Marantz in The New Yorker, the “overriding impression” of Altman is that he is a “borderline sociopath”, said Jeremy Kahn in Fortune. The piece raises questions on whether Altman “actually cares about AI safety” or whether his rhetoric is simply a “convenient pose” to win over funders and regulators.

Dario Amodei

“We should not deny that the disruption is going to happen” as AI use increases, Anthropic CEO Amodei told John Thornhill in the FT, but AI can only “diffuse at the speed of trust”. Trust, however, said Thornhill, is “in short supply”. “As the current frontrunner of the AI pack, Amodei is certain to come under increasingly fierce scrutiny.”

It is clear that he “wants to position himself as one of the good guys in the AI debate”, but that “grates with many Silicon Valley critics”, who argue that “his principles align with Anthropic’s commercial interests”. Amodei founded Anthropic – the creators of Claude – in 2021 alongside six other former OpenAI employees, including his sister Daniela, who is president. The company has recently raised $30 billion (£22.2 billion) at a $380 billion (£281.3 billion) valuation and is reportedly “heading for a giant stock market flotation later this year”.

Central to Amodei’s brand of Anthropic is that it is “fundamentally safer than that of its rivals”, said The Wall Street Journal. Indeed, that was one of the main reasons Amodei left OpenAI, citing “concerns about safety”. In recent months, he has also “compared the legal battle between Altman and Elon Musk to the fight between Hitler and Stalin”, as well as calling a $25 million (£18.5 million) donation by OpenAI President Greg Brockman to pro-Trump super PAC (independent expenditure-only political action committee) Maga Inc. “evil”.

Jensen Huang

Although the head of Nvidia may not be driving the AI revolution directly, his company is facilitating it, acting as the “hardware backbone” of the movement, said Business Insider. Huang’s “chip empire” is effectively “powering the generative AI boom”.

He founded the company in 1993, and has served as CEO ever since. Under his leadership, Nvidia – whose projected revenue opportunity for its artificial intelligence chips could reach $1 trillion (£740 billion) or more by the end of 2027 – has expanded partnerships with Amazon Web Services, Google Cloud and Oracle Cloud to accelerate AI development. Nvidia’s hardware and software “now sit at the centre of nearly every major foundation-model program”, said Business Insider.

AI is “gonna create more jobs in the end”, Huang said during a recent panel at the Stanford Graduate School of Business, reported Fortune. “There’ll be more people working at the end of this industrial revolution than at the beginning of it.” He has previously commented that negative commentary surrounding AI is “extremely hurtful”, said Inc.

Huang is not without his quirks, having banned one-on-one meetings with staff who report directly to him, on the grounds they would “clog up his work schedule and slow him down”, said Fortune.

Alex Karp

Fewer people will have heard of the co-founder of Palantir, but to some he is the “scariest CEO in the world”, said Steve Rose in The Guardian.

The company recently released a 22-point “manifesto” summarising Karp’s recent book, “The Technological Republic: Hard Power, Soft Belief, and the Future of the West”. In it, he extols the need for “hard power”, argues the inevitability of “AI weapons” and calls for the reversal of the “postwar neutering of Germany and Japan”. MPs have since called this a “parody of a ‘RoboCop’ film” and the “ramblings of a supervillain”. Arguably, what it does show is that “Karp views himself as not simply the head of a software company, but a pundit with important insights into the future of civilisation”, said Aisha Down and Robert Booth in The Guardian.

The company is “at the heart of many of the world’s pressing issues”, said The Guardian. Palantir has “multibillion-dollar contracts” with the US Army and Ice, as well as partnerships with the Israeli military and the Ministry of Defence, said Al Jazeera.

Some NHS staff are “refusing to work” on the health service’s Federated Data Platform, which is provided by Palantir, due to the company’s “role in US defence and immigration enforcement”, said the FT. Ministers are exploring the possibility of a “break clause” in the company’s seven-year £330 million NHS contract, signed in 2023.

Elon Musk

The founder of xAI and Grok, such is the strength of Musk’s conviction in AI, that he believes it will put “immortality within human reach”, said Fortune.

But the “rapid rise” of his tech company xAI’s has “raised concerns”, said Harry Booth in Time. There were accusations of pollution from the Colossus data centres’ temporary gas turbines, and the now-infamous update to Grok “praised Adolf Hitler as a ‘decisive leader’ and began creating graphic rape narratives”.

French prosecutors summoned Musk for a voluntary interview on Monday, which he did not attend, over “alleged abuse of algorithms and fraudulent data extraction” by his AI chatbot Grok, as well as the “creation of sexual deepfakes”, said France 24. This is part of an ongoing probe first opened in 2025, with the company’s offices raided by the Paris prosecutor’s cybercrime unit in February.

Musk is also locked in a legal feud with Altman – with whom he cofounded OpenAI – accusing Altman of deceiving him into donating $38 million (£28 million) towards the company with the promise that it would remain a non-profit, said Business Insider.

It was not a joke. The shoe company Allbirds announced last week that it is pivoting to artificial intelligence, a sign that the AI bubble is about to pop. Or maybe the tech optimists are right and everything is AI now.

The company was “once the maker of Silicon Valley’s favorite shoe,” said The New York Times. Allbirds was previously valued at $4 billion, but the company earlier this year closed all its stores and sold its assets for a mere $39 million. Now the brand seeks a fresh start: The business is rebranding itself “NewBird AI” and announced it had received a $50 million influx to buy up advanced computer chips that will let it enter the AI infrastructure business. That investment is a “drop in the bucket” for an industry spending billions to build data centers, but Wall Street loved the news. NewBird’s stock immediately rose nearly 600%.

The market’s reaction proves “AI excitement is alive and well — but as silly as ever,” Noah Weidner said at The Street. The move might make sense, though. Artificial intelligence requires a “massive volume” of computing power, and companies able to furnish it “will drum up excitement” — even if that company once sold shoes.

AI is creating wealth

Will AI spending hold up?

The shoe company’s “flailing AI embrace” is “not a horrible idea on the surface” given that it fills a “real business need,” Nitish Pahwa said at Slate. But the AI spending that has “propped up the economy” might not persevere, and communities are “successfully obstructing the data centers” needed for further expansion. Indeed, Allbirds’ stock started to drop after the initial surge, said Bloomberg. The market roller coaster ride gives Allbirds the feel of a “meme stock,” said 50 Park Investments’ Adam Sarhan, in which “emotions take over and logic and reason get thrown out the window.”

What happened

A humanoid robot called Lightning won a half-marathon in Beijing on Sunday, beating his robotic competitors and the human runners in a parallel race by completing the 13-mile course in 50 minutes and 26 seconds — nearly seven minutes faster than the world record set by Uganda’s Jacob Kiplimo in Lisbon last month.

Who said what

The victory of Lightning, built by Chinese smartphone brand Honor, marked a “significant step forward from last year’s inaugural race,” when the winning robot “finished in 2 hours, 40 minutes and 42 seconds,” The Associated Press said. The “remarkable feat” was also a “big stride for China in its technological rivalry with the U.S.,” CNN said.

China already has “more robots at work” than “the rest of the world combined,” The New York Times said. Beijing also recently hosted the first Humanoid Robot Games, featuring “plenty of running, kicking and punching,” though the robots “also flailed around, crashed and fell over many times.”

What next?

The leap forward in China’s humanoid engineering “is genuinely impressive,” Oregon State University robotics professor Alan Fern told the Times. But it’s “much less obvious” how a robot winning a half-marathon “translates into productivity and ultimately, profitability.”

The Internet Archive has been responsible for saving and providing access to trillions of websites over the past 30 years. AI is putting a damper on the organization’s work, as large language models are using the data without permission. As a result, many companies are no longer allowing their content to be archived, which could lead to a large loss of historical records in the future.

Access denied

The Internet Archive is a non-profit that is building a “digital library of internet sites and other cultural artifacts,” according to its website. The organization uses web crawlers to capture snapshots of sites. These snapshots are then made available through the public-facing tool, the Wayback Machine, which operates like a library, providing “free access to researchers, historians, scholars, people with print disabilities and the general public.” However, amid the rise of AI, the Internet Archive’s “commitment to free information access has turned its digital library into a potential liability for some news publishers,” said an analysis by Nieman Lab.

Currently, “241 news sites from nine countries explicitly disallow at least one out of the four Internet Archive crawling bots,” including The New York Times and Reddit, said Nieman Lab. Of these sites, 87% are owned by USA Today Co., the “largest newspaper conglomerate in the United States, formerly known as Gannett.” The Guardian has also restricted the Internet Archive; the publication does not block the crawlers, but it “excludes its content from the Internet Archive API and filters out articles from the Wayback Machine interface, which makes it harder for regular people to access archived versions of its articles,” said Wired.

Many of the same media outlets banning Internet Archive’s crawlers have used the resource themselves to access older data and articles. “Journalists rely on the Archive as a resource in our reporting, and many digital investigations into issues like misinformation or censorship are possible only because it preserves material that would otherwise disappear,” said the organizations Fight for the Future, the Electronic Frontier Foundation and Public Knowledge, in a letter to the Internet Archive. “Without that ongoing work to preserve the web, large parts of journalism’s recent history would already be lost.”

On record

Artificial intelligence is the biggest reason sites are blocking the Internet Archive. There is “evidence that the Wayback Machine has been used to train large language models,” said Forbes. The archive allows tech companies to “skirt copyright laws by using the Wayback Machine as a workaround for training language models on their content,” said Morning Brew. Despite this, Mark Graham, the director of the Wayback Machine, “emphasizes that the digital archive has controls to limit abuse of AI automation and prevent large-scale data extraction.”

Unfortunately, a few bad apples ruin the whole bunch. The Internet Archive “tends to be good citizens,” Robert Hahn, the head of business affairs and licensing at The Guardian, said to Nieman Lab. “It’s the law of unintended consequences: You do something for really good purposes, and it gets abused.” The nonprofit “has taken on the Herculean task of preserving the internet, and many news organizations aren’t equipped to save their own work,” Nieman Lab said.

There is “no widely available public tool comparable to the Wayback Machine,” said Wired. If it “continues to lose access to major news sources, its preservation efforts could erode to the point where early digital records of history become much harder to access or are even lost altogether.”

Appearing straight out of science fiction, portable computers called cyberdecks have been growing in popularity, especially among Gen Z. They can be built with minimal parts and customized both in their purpose and aesthetic. The trend is a response to a perceived lack of creativity in mainstream technology, as well as a way to fight back against data harvesting. And many want to use technology without the influence of large corporations, similar to the days of the early internet.

From mind to machine

A cyberdeck is a transportable, homemade computer “used to access an online interface,” said Daily Dot. The term originated with the 1984 sci-fi novel “Neuromancer” by William Gibson. And since then, cyberdecks have been a “staple of the cyberpunk genre and aesthetic.” Building them has become a trend among Gen Z, blending “retro-futuristic aesthetics with practical computing,” said Newsweek.

Cyberdecks are generally simple to construct, often using “single-board systems like Raspberry Pi paired with small screens, keyboards and custom enclosures,” said Newsweek. Many are also “built from thrifted or repurposed materials, giving each device a distinct look and function shaped entirely by its creator.”

These hand-built computers serve a variety of purposes, including gaming machine, e-reader, information database or MP3 player. And building a cyberdeck “can be as complex or simple as you choose to make it,” said The Cyberdeck Cafe. “People of all skill levels have constructed their own.”

Cyberdecks are “open systems, meaning components can be swapped, modified or redesigned,” said Newsweek. The flexibility is “part of the appeal for younger users who want to experiment with hardware and software without restrictions.”

The trend comes at a time when technology and social media platforms have become controlling with “more data harvesting, more algorithmic control, more ads, more surveillance,” said Quasa. Cyberdecks are “less about replacing everyday devices and more about reclaiming control over technology,” said Newsweek.

Sticking it to the man

Building a portable computer is a “way fringe and anti-establishment engineers and cyberpunks are creating a digital identity all their own,” said Daily Dot. Cyberdecks “combat the unbounded corporatization, invasiveness and homogeneity of widespread tech, in addition to individualizing the tech experience according to a user’s aesthetic.”

They are “quietly rebellious” and a “direct middle finger to the boring, minimalist ‘everything-is-a-sleek-black-rectangle’ aesthetic that dominates tech design,” said Quasa. Much of the love for cyberdecks is a result of disillusionment with the state of modern technology. The “early internet’s wild, private, joyful chaos feels like a distant memory.” Gone is the world in which “you didn’t chase likes or dread the next feed update.”

While technology has been “shaping the world’s digital future,” cyberdecks are “driving users back to the past — a time when a simpler, less corporatized and aggressively monitored online reality once existed,” said Daily Dot. The trend is “self-defense and nostalgia at the same time,” said Quasa. “When you are making something that’s truly yours, why be boring? Make it fun. Make it ridiculous. Make it you.”

As part of AI company Anthropic’s Project Glasswing initiative, the new general-purpose model Mythos is uniquely powerful in the artificial intelligence industry and is causing concern among even people who are normally trusting of AI. The company, which also makes the AI model Claude, has claimed that Mythos is currently too advanced for public release, and is instead entrusting the model to cybersecurity experts for the time being. Some are worried this could pave the way for even more nefariousness in the AI space.

‘New era of hacking’

Mythos’ AI programming is able to find potential weaknesses in cybersecurity, and it can “detect thousands of high- and critical-severity bugs and software defects, with vulnerabilities identified in most major operating systems and web browsers,” said NBC News. Some of these vulnerabilities “had been undiscovered for decades,” according to Anthropic’s experts. The company found that Mythos’ “cybersecurity capabilities in particular were surprisingly advanced” compared to similar general-purpose AI models.

But there are also fears that Mythos “could usher in a new era of hacking and cybersecurity,” said NBC News. Mythos is “capable of advanced reasoning,” which could allow it to “identify and exploit a growing number of software vulnerabilities” if it were to fall into the wrong hands. To stave off these fears, Anthropic is allowing certain tech firms to access Mythos. But the company “does not have plans yet to release Mythos to the general public,“ said Bloomberg, a move that will ensure the AI ends up “in the hands of defenders first,” officials with Anthropic said.

The tech firms are expected to use Mythos as part of a project called Glasswing to “hunt for flaws in their products and share findings with industry peers,” said Bloomberg. It is a notable change because it will be the “first time a leading AI lab has built a frontier model and simultaneously decided the public cannot use it,” said Forbes. Anthropic’s position remains “straightforward: The model’s cyber capabilities are too dangerous for general availability.”

‘Humanity’s most devious behaviors’

In addittion to hacking vulnerabilities, some experts are concerned about Mythos’ capabilities. Anthropic released a safety evaluation for Mythos that shows a “striking leap in scores on many evaluation benchmarks,” the company said. In some instances, the evaluation “reads like a thriller about an AI that has learned some of humanity’s most devious behaviors,” said Axios.

At least one of the tests performed by Anthropic showed Mythos “acting like a cutthroat executive,” said Axios, doing things like “turning a competitor into a dependent wholesale customer, threatening to cut off supply to control pricing and keeping extra supplier shipments it hadn’t paid for.” The AI had instances where it “used a prohibited method to get an answer, then tried to ‘re-solve’ it to avoid detection,” though these were limited to “less than 0.001% of interactions.”

These issues have not stopped companies from working with Mythos, as “approximately 40 organizations involved in the design, maintenance or operation of computer systems are said to have joined Glasswing,” said The Guardian. This includes major firms like Amazon, Apple, Google, JPMorganChase and Microsoft. And while Anthropic has previously sparred with the Trump administration about its implementation in the Defense Department, the company has also “had discussions with the U.S. government regarding Mythos.”

Americans may be unwittingly giving hackers an easy path to access their houses. Cybersecurity experts, including FBI cybercrime analysts, are warning about residential proxy networks found on many off-brand electronics. These networks often allow hackers to hide in plain sight.

What are residential proxy networks?

These software systems are “designed to route other people’s internet traffic through a user’s device,” said Cyber Magazine. The networks operate largely like “forged return addresses on envelopes — someone else’s internet traffic is rerouted through your connection,” said officials at Comcast’s Threat Research Lab to Cyber. As the networks engage with users, they “quietly launder illegitimate activity” while making it appear that your device is the “initiator of that traffic.”

Residential proxy networks can make their way onto a variety of home devices, as “TV streaming devices, digital picture frames, smartphones, tablets and routers are used to route traffic,” said the FBI. Many people who own such devices do not “realize their internet connection could be used by someone else without their permission.” The devices can sometimes gain internet access when the “owner of the device provides consent” unintentionally; other times, the owner “does not provide consent and is unaware their IP address is being used.”

Some of these devices “ship with residential proxy software preinstalled on them,” which can “happen with certain low-cost video streaming systems,” said The Wall Street Journal. In other cases, people might “download the code to their smartphones” without realizing it. And since the networks make it appear like illegal activity is coming from an innocent person’s home, there’s a “chance that law enforcement could come knocking at your door.”

How can people protect themselves?

The FBI has a list of tips to help people stay safe, urging Americans to “avoid TV streaming devices that claim to provide free sports, TV shows and movies,” as these “may contain malware or backdoors that hijack your internet network and can lead to identity theft,” said the agency. The agency also recommended people be wary of downloading free VPNs and clicking on pop-ups, which can “initiate malware installation on your device.”

In the midst of these continued cyberattacks, some ordinary Americans are fighting back. Benjamin Brundage, a senior at the Rochester Institute of Technology, began an investigation in 2025 as a “growing network of hacked devices was launching the biggest cyberattacks ever seen on the internet” via a Chinese company called Ipidea, said the Journal. Using cat memes to “lighten the mood” while speaking to hackers, Brundage was able to find out significant information about the attackers, and law enforcement eventually “took action against the network.”

Brundage “identified 11 of the largest residential proxy companies, including Ipidea, that were vulnerable” to hackers, said the Journal. Other companies also assisted law enforcement in the investigation. Google “took legal action” against Ipidea to “take down domains used to control devices and proxy traffic through them,” said the tech company in a press release. While there are still “significant challenges for network defenders to detect and block malicious activities,” officials believe the action taken against Ipidea has reduced the “available pool of devices for the proxy operators by millions.”

It’s no secret that artificial intelligence can sometimes offer less-than-stellar guidance. But AI might give people this bad wisdom for a sobering reason: to flatter, according to a new study. In some cases, AI may only reinforce people’s preconceived notions, but the words it generates can be outright harmful.

What did the study find?

The “sycophantic (flattering, people-pleasing, affirming) behavior” of AI chatbots can pose risks as people “increasingly seek advice about interpersonal dilemmas,” said the study published in the journal Science. In an analysis of 11 leading large language models, including AI bots from Anthropic, Google and OpenAI, chatbot responses to users were “nearly 50% more sycophantic than humans’, even when users engaged in unethical, illegal” behaviors.

The problem is not just that these chatbots “dispense inappropriate advice but that people trust and prefer AI more when the chatbots are justifying their convictions,” said The Associated Press. In one example, when OpenAI’s ChatGPT was asked if littering in a park was acceptable if no garbage can was available, the bot “blamed the park for not having trash cans, not the questioning litterer who was ‘commendable’ for even looking for one.”

This example may seem trivial, but AI’s general tendency to “flatter and excessively confirm users’ opinions can lead to wrong decisions, harm relationships and reinforce harmful beliefs while decreasing the willingness to take responsibility or resolve conflicts,” said The Jerusalem Post. The proneness toward sycophancy is a “technological flaw already tied to some high-profile cases of delusional and suicidal behavior in vulnerable populations,” said the AP.

Why is AI sycophancy such a problem?

Many experts worry that this AI advice “will worsen people’s social skills and ability to navigate uncomfortable situations,” Myra Cheng, the study’s lead author and a computer science PhD candidate, said to the Stanford Report. If this behavior by AI is not corrected, some users may “lose the skills to deal with difficult social situations” and could also pose larger safety risks.

“Users are aware that models behave in sycophantic and flattering ways,” Dan Jurafsky, the study’s senior author and a Stanford University linguistics professor, told the Stanford Report. What many people are “not aware of, and what surprised us, is that sycophancy is making them more self-centered, more morally dogmatic.” This type of interaction with AI is a “safety issue, and like other safety issues, it needs regulation and oversight.” All of this is also happening as AI use becomes more prevalent, especially among teenagers.

At least 33% of teens “use AI companions for social interaction and relationships, including conversation practice, emotional support, role-playing, friendship or romantic interactions,” according to a study from Common Sense Media. Another 33% of teens choose to “discuss important or serious matters with AI companions instead of real people.” Experts say when using AI you should avoid asking for advice on crucially important topics. “I think that you should not use AI as a substitute for people for these kinds of things,” Cheng told the Stanford Report. “That’s the best thing to do for now.”

‘When capital can think, who pays?’

Ravi Kumar S, Andreea Roberts and Simone Crymes at Newsweek

In the U.S., AI adoption is “growing at a remarkable pace,” but Americans are “concerned” about “layoffs tied to automation,” say Ravi Kumar S, Andreea Roberts and Simone Crymes. So how should “public policy support” the transition? One answer: a “shift in how automation is taxed relative to human labor.” If capital is “taxed more and labor less, replacing people with AI is no longer the cheapest path,” and using AI to “augment human workers” instead “becomes a more attractive option.”

Read more

‘Birthright citizenship made me American. We can't lose it.’

Cynthia Choi at USA Today

On his “first day back in office,” Trump issued an executive order “seeking to deny citizenship to certain U.S.-born children,” says Cynthia Choi. But birthright citizenship is as “fundamental” to our country as “freedom of speech.” This is “not some isolated policy debate.” It’s a “broader effort by the Trump administration to put an end to multiracial democracy.” Children without citizenship will be denied “access to education, public benefits and the basic rights that come with belonging.”

Read more

‘Missile warfare is faster, deadlier and harder to control’

Hal Brands at Bloomberg

The Iran conflict “demonstrates how the spread of powerful, accurate missiles is changing warfare around the globe,” says Hal Brands. Even “relatively weak states now have fairly accurate weapons that can strike hundreds, even thousands, of miles away.” This means “fewer sanctuaries: Facilities and geographies that were once secure are now vulnerable to attack.” That could be “challenging” for the U.S., since “even relatively weak adversaries will be able to hold U.S. bases, perhaps even the homeland, at risk.”

Read more

‘AOC finally takes a position that makes sense on military aid to Israel’

Zeeshan Aleem at MS Now

On Tuesday, Rep. Alexandria Ocasio-Cortez (D-N.Y.), who “struggled to take a clear position on supporting Israel in the past,” pledged to vote “against all military aid to Israel,” says Zeeshan Aleem. This was a “striking shift for a potential 2028 White House hopeful who, should she enter the race, would be the standard bearer for the democratic socialist left.” Her decision “does not just reflect demands on the left but the changing dynamics of the Democratic Party.”

Read more

“If you look backward in this business, you’ll be crushed. You have to look forward,” said Apple co-founder and CEO Steve Jobs in 2008, a year after he introduced the first iPhone and changed the world forever.

Apple may indeed be “allergic to nostalgia”, said Steven Levy in Wired, but the company is still “begrudgingly engaging in a series of concerts and commemorations, and we’re being blitzed by books, articles and oral histories” to mark its 50th anniversary.

From an inauspicious start in Jobs’ California garage, the company he founded with Steve Wozniak in 1976 went on to pioneer the personal computer, transform the music market, and revolutionise how people use technology in the internet age. Apple is now valued at more than $3.6 trillion (£2.7 trillion), generating $400 billion (£301 billion) a year in revenue, with iPhone sales alone expected to bring in $1 million (£750 million) every 90 seconds. Across the world, 27% of the population – roughly 2.2 billion people – use one or more of its products.

Tariffs, trade wars and anti-trust trials

“No country has been more central to Apple’s rise – or more fraught for its future – than China,” said France 24. CEO Tim Cook, who took over from Jobs after he died of pancreatic cancer in 2011, made China the primary manufacturing base for Apple devices. It is also one of Apple’s largest consumer markets but the company “faces mounting pressure” from “trade tensions and tariffs” accelerating efforts to diversify manufacturing elsewhere in Asia, while “competition from domestic rivals such as Huawei has eaten into Apple’s Chinese market share”.

To put it bluntly, “the world in which Apple once thrived no longer exists,” said former Financial Times editor Lionel Barber in The New Statesman. A “25-year-long process of hyper-globalisation in which money, technologies and ideas have flowed freely” is “now fading amid economic nationalism driven, in part, by a technological arms race between the US and China, and a global tariff offensive led by Donald Trump”.

Apple is also facing a threat to its dominance closer to home, in the form of a series of anti-trust cases against it. “In an industry full of sprawling multipronged tech empires”, the basic argument against Apple is “comparatively simple”, said Adi Robertson on The Verge: “it’s become the ultimate gatekeeper to billions of people’s primary computing hardware, and it keeps competitors locked out while levying a heavy toll on the developers it lets through”.

Regulators and courts have ordered changes, particularly around the App Store, “but those changes have been slow to arrive, in part because for a half-decade or more, Apple has dragged its feet at every turn”.

Artificial intelligence

Apple may have “absolutely owned” the internet and mobile era, said Wired, but “now the future belongs to AI” – a category where Apple seems to have been lacking.

Apple’s Siri lags behind the likes of Microsoft, OpenAI’s ChatGPT, Anthropic’s Claude and Google’s Gemini, let alone China’s DeepSeek.

This is, in part, because Apple is “limited by its ecosystem”, said Acuity Trading. AI systems “require vast amounts of data, public testing and continuous version launches” and so “cannot be perfected in a closed ecosystem, which is what Apple has built its reputation on”. But perhaps the “most limiting factor is that Apple takes its commitment to user privacy very seriously”, which “has hindered AI development by limiting the amount of data it can use for training AI models”.

This “obsession with user privacy and its premium hardware could position it to drive widespread adoption of personalised AI – and make it profitable, a goal that has proved elusive for much of the AI industry”, said France 24.

Succession planning

The demise of Apple has been predicted many times before; in the mid-1980s after Jobs was forced out and again in 2011 when he passed away. Having revived the company and driven the release of the iMac, iPod and iPhone, Jobs was “widely thought of as irreplaceable”, said Barber. But Cook has not only steadied the ship but also taken the company to new heights, in terms of revenue generation if not technological innovation.

While the 65-year-old has given no indication of an imminent transition, the most likely candidate to take over when he does decide to go is John Ternus, senior vice president for hardware engineering, who oversees development of the devices that generate roughly 80% of Apple’s revenue. “Known for his steadiness and political acumen”, Ternus, like Cook, is “risk averse” and would be a continuity hire rather than “someone more willing to shake things up”, said Bloomberg.

This matters because, while its products “helped define the past 50 years of consumer technology, thriving for another 50 will inevitably require the company to transform in ways that aren’t entirely clear today”.

This week saw what could prove to be an historic reckoning for Big Tech when a Californian court ruled that Meta and Google’s YouTube intentionally built addictive social media platforms. This came just a day after a jury in New Mexico found Meta liable for the way its platforms endanger children.

Critics are calling this “Big Tech’s Big Tobacco moment”, a reference to how cigarette makers in the 1990s had to overhaul their businesses after courts ruled that their products were addictive and harmful.

Meta and Google have invested heavily in safety tools for younger users and both companies dispute claims that their platforms are to blame for children’s mental health issues. But the verdicts this week are a “sombre moment for Silicon Valley and the implications are global”, said the BBC’s technology editor Zoe Kleinman.

What did the commentators say?

The cases this week are “the first of about 22 ‘bellwether’ trials”, said Stephen Armstrong in The Independent, brought by more than 350 families across 250 US school districts and are “expected to trigger thousands more”. It is like the “anti-tobacco legal actions on fast-forward”.

Judgments of responsibility “in cases like the one brought against Meta and YouTube are necessarily complex”, said academic and author Austin Sarat in The Guardian. And critics of the judgment in this case “will no doubt howl about greedy plaintiffs looking to make a haul from deep-pocketed defendants”. But it does seem “clear that companies knew of the addictive qualities of their sites and the potential damage to young people”.

For years, “technology giants successfully fought off efforts by regulators, lawmakers and others to put limits on their social media businesses”, said Andrew Ross Sorkin in The New York Times. The tide appears to be turning but so far “investors don’t seem to be fazed”, with Big Tech share prices only dipping slightly. The potential penalties too – $6 million for Meta and YouTube in California, and $375 million for Meta in New Mexico – “are a fraction of their immense profits”.

It’s for that reason that social media companies might not fret too much too soon. “The Big Tech firms are losing nearly every time,” Tom Smith, partner at legal firm Geradin, told The Independent. “But they have effectively unlimited legal budgets, and their calculation may be that as long as you can make sure these cases take a decade, then the extra profits will outweigh the damages.”

What next?

Meta and YouTube plan to appeal, but if unsuccessful “they could be forced to remove the features that make their platforms addictive, which would upend their business models and fundamentally alter the experience of users”, said Fred Harter in The Observer.

Regardless of whether Meta or Google appeal the decision, “this is going to redefine the landscape,” said the BBC’s Kleinman. “It could even be the beginning of the end of the social media era as we know it.”

What happened

A New Mexico jury on Tuesday found that Meta knowingly harmed children’s mental health, enabled child sexual exploitation and misled users about the safety of its Instagram, Facebook and WhatsApp platforms. The jurors ordered Meta to pay $375 million in civil penalties for thousands of violations of the state’s Unfair Practices Act. It was the first major courtroom loss for Meta in a growing number of lawsuits accusing it and other social media giants of harming or failing to protect young people.

Who said what

The verdict is a “historic victory for every child and family who has paid the price for Meta’s choice to put profits over kids’ safety,” said a statement from New Mexico Attorney General Raúl Torrez, who filed the suit in 2023. During the seven-week trial, the jurors “were presented with internal Meta documents and heard testimony from former employees,” the BBC said, including whistleblower Arturo Béjar, who testified that Meta was aware underage users were being served sexualized content and “said his own young daughter was propositioned for sex by a stranger on Instagram.”

The jurors “ordered a maximum penalty for each violation,” The Wall Street Journal said, but “Meta made 160 times” the $375 million fine “in its most recent quarter” alone. Shares of the $1.5 trillion company were “up 5% in early after-hours trading following the verdict,” The Associated Press said. Still, the “landmark decision” in Santa Fe “signals a changing tide against tech companies and the government’s willingness to crack down” on social media’s harms to young people.

“Parents, policymakers and the tech industry watched the New Mexico case closely for its potential to force Meta to change the design of its products,” The New York Times said. But the trial judge, not the jury, will rule on any compulsory changes for Meta at the trial’s second stage in May.

What next?

“We will continue to defend ourselves vigorously,” said Meta spokesperson Andy Stone, and “we remain confident in our record of protecting teens online.” A jury in Los Angeles “has been deliberating for more than a week” in a separate “bellwether social media addiction trial” accusing Meta and YouTube of “harming the mental health of a user through addictive design features,” the Times said.

Social media posts appeared last month calling for “red v blue” wars between schools, but instead of provoking fights between students, the posts appear to have made a deeper impact on their worried parents, leading experts to suggest practising an online strategy known as critical ignoring.

It’s a concept that experts are “increasingly teaching”, Sander Van Der Linden, a professor of social psychology, told the BBC, and it “will become more important in the face of AI-generated slop, where sometimes it’s better to just ignore low-quality stuff”.

What is critical ignoring?

It’s a behavioural strategy for managing information overload by consciously choosing to filter out low-quality, distracting, or manipulative content. People look for clues that allow them to ignore a post. While critical thinking analyses information, critical ignoring decides what to analyse in the first place, serving as a crucial filter.

Critical thinking is not enough “in a world of information overabundance and gushing sources of disinformation”, wrote Ralph Hertwig, Anastasia Kozyreva, Sam Wineburg and Stephan Lewandowsky on The Conversation.

The digital world “contains more information than the world’s libraries combined”, so “critically thinking through all information and sources we come across” would “utterly paralyse us”.

Also, “investing critical thinking in sources that should have been ignored in the first place” results in “attention merchants and malicious actors” getting what they wanted: “our attention”.

Doesn’t AI help with this?

To an extent. AI chatbots can help people understand what’s true and untrue on the internet, but they are tools, rather than perfect judges of truth.

ChatGPT has “introduced a new temptation” – the “feeling that I can get a clean answer to everything, instantly”, said Tom’s Guide. But this is where things “get tricky” because ChatGPT is “so fluent, so confident, so fast, it can make ‘done’ feel like ‘true’”, and “‘sounds right’ feel like ‘verified’ – even when it’s not”.

So it’s “up to us, as individuals, to stop ingesting the pink slime of AI slop, the forever chemicals of outrage bait and the microplastics of misinformation-for-profit”, said The Wall Street Journal. Critical ignoring is a widely recommended strategy for this.

But how do I do it?

The “key word” is “critical”, said Psychology Today, because it doesn’t mean “just ignoring everything”. Rather you should look quickly for clues that suggest the types of information most likely to be misinformation or disinformation.

The clues include signs that it’s polarising content, that it “appeals to intuition or common sense”, instead of “including facts or evidence”. Another red flag is if it doesn’t include sources, or those it does don’t seem credible. Does it seem to have been released “as a distraction”, or does it promote “the threat of a bogeyman or a scapegoat”?

Then there’s “lateral reading”, a more time-consuming strategy which “involves opening up new browser tabs to search for information” about the “organisation or individual behind a site” before “diving into its contents”, said The Conversation. Also, it’s always a good idea to not “feed the trolls”.

“Remember that your attention is a scarce resource”, said The Wall Street Journal, and “decide how much time you want to spend on screens in advance, then set a timer.”

A practice called “self-nudging” includes removing “distracting and irresistible notifications”, or setting “specific times in which messages can be received”, thus “creating pockets of time for concentrated work or socialising”, said The Conversation.

Or you can just “ask one question“ before engaging, said Tom’s Guide. “Would I care about this tomorrow?” If not, you can simply “move on”.

What happened

Anthropic on Monday sued Defense Secretary Pete Hegseth, the Defense Department and several other federal agencies in federal court, arguing that the administration’s move to blacklist the AI firm as a national security risk was “unprecedented and unlawful.” The Constitution “‌does not allow the government to wield its enormous power to punish a company for its protected speech,” Anthropic said in its filing.

Hegseth last week formally designated the company a “supply chain risk” over Anthropic’s insistence that its AI tool Claude not be used for autonomous lethal weapons or mass surveillance of Americans. President Donald Trump said on social media that all federal agencies must stop using Claude within six months.

Who said what

The supply-chain risk designation “effectively cuts off Anthropic’s work with the Defense Department” and its contractors, The New York Times said, and it “has never been used on an American company.” The label is “usually reserved for Chinese and Russian firms suspected of helping foreign spies,” The Washington Post said. The Pentagon’s “unprecedented step” came “even as Anthropic’s tools were playing a central role” in “Trump’s bombing campaign in Iran.”

“It is absurd for the government to argue that Anthropic is the kind of company meant to be addressed by this statute,” especially when the Pentagon “has repeatedly sought to obtain Anthropic’s services for national defense,” Georgetown University law professor Mark Jia told the Post. It would be “perfectly reasonable” for the Pentagon to cancel its contracts with Anthropic because they don’t believe a private company should set policy or determine when “autonomous lethal weapons are ready for prime time,” Dean Ball, a former Trump White House AI policy adviser, said on “The Ezra Klein Show.” But they don’t have the “statutory power” to “completely destroy the company” in “a kind of political assassination.”

What next?

The White House is “preparing an executive order formally instructing the federal government to rip out Anthropic’s AI from its operations,” Axios said, and it “could be issued as soon as this week.” Anthropic’s “standoff with the Defense Department has cost it Uncle Sam as a customer,” The Wall Street Journal said, but it has also brought a “surge of public goodwill” and a “momentary advantage in the ferocious talent war between rival artificial intelligence labs.”

What does France’s nuclear policy shift mean for Europe? Why is the dinosaur market booming? And can North Korea regain its place at the top of women’s football?

Olly Mann and The Week delve behind the headlines and debate what really matters from the past seven days.

A podcast for curious, open-minded people, The Week Unwrapped delivers fresh perspectives on politics, culture, technology and business. It makes for a lively, enlightening discussion, ranging from the serious to the offbeat. Previous topics have included whether solar engineering could refreeze the Arctic, why funerals are going out of fashion, and what kind of art you can use to pay your tax bill.

You can subscribe to The Week Unwrapped wherever you get your podcasts:

• Spotify
• Apple Podcasts
• Global Player

The Trump administration has long trumpeted its goal to automate its operational capacity through artificial intelligence models provided by companies like OpenAI and Elon Musk’s xAI. But as Secretary of Defense Pete Hegseth moves to offload certain human operations into the realm of the algorithm, one tech firm has emerged as a counterbalance to the White House’s vision for an artificially intelligent military: Anthropic, which “cannot in good conscience” allow Hegseth’s Pentagon to use its AI models without limitations, said CEO Dario Amodei. As the Defense Department weighs consequences, other AI firms are starting to take note — and weigh in.

Taking a ‘bold stand on ethical grounds’

Despite believing in the “existential importance” of using AI to protect the United States and “defeat our autocratic adversaries,” Anthropic has identified a “narrow set of cases” including mass domestic surveillance and “fully autonomous weapons” wherein AI can “undermine, rather than defend, democratic values,” Amodei said in a company statement. Moreover, Hegseth’s allegedly retaliatory move to blacklist Anthropic as a supply chain risk is "inherently contradictory” for labeling the company a security risk and simultaneously “essential to national security.” Hegseth's “heaviest-handed way you can regulate a business” marks a “landmark moment” for how the Pentagon “interacts with our cutting-edge technology developed on U.S. soil” in general, said Katie Sweeten, a former Justice Department official who coordinated the relationship between DOJ and the Pentagon, to Politico.

While Amodei's Anthropic faces a government ban, his “main rival,” OpenAI's Sam Altman, "struck his own deal” to fill Anthropic's Defense Department role, said CBS News. Reached just hours before the U.S. and Israel launched a joint assault on Iran, the OpenAI partnership did not prevent the military from using Anthropic's “very same tools” that it had just banned, said The Wall Street Journal. It will likely take “months” to fully replace Anthropic’s Claude AI model with other platforms.

By “refusing to bow” to a White House intent on “bullying private companies into submission,” Amodei is “taking a bold stand on ethical grounds,” said The Atlantic. While the company’s competitors “jockey for dominance” in the field, Anthropic has “distinguished itself by emphasizing safety.” Refusing White House pressure means Anthropic “may have just averted another crisis” in the form of a “major public backlash” from those who could see the company as a “more principled player in the AI wars.” After Altman's OpenAI replaced Anthropic at the Pentagon, the latter's Claude app has been "rocketing to the top of the App Store,” with some users saying they were “defecting” from ChatGPT to Anthropic after feeling “uneasy about OpenAI's ambitions," said Business Insider.

Contract negotiation vs. congressional regulation

Anthropic is “rightly concerned” that its products could be used for “unsafe or malicious” ends, said former Air Force Secretary Frank Kendall at The New York Times. But the company is wrong for trying to use “contractual terms” to either “prevent the misuse of its products,” or at least to “deflect responsibility.” But Anthropic also has the “option” to not sell to the government at all. The government, meanwhile, “cannot be expected to negotiate provisions” like Anthropic is asking for with all its partners, which would be a “nightmare to administer and unenforceable.” What, then, could be “appropriate” to address this debate? “Regulation by Congress.”

While it’s no secret that artificial intelligence can be used for nefarious purposes, China is working with AI on an unprecedented scale and using it to target its enemies, according to a report from OpenAI. This is just one way that China is employing AI behind the scenes to disrupt global operations.

‘Sprawling Chinese influence operation’

The report claims that China is using OpenAI’s chatbot, ChatGPT, as a “diary to document the alleged covert campaign of suppression,” said CNN. A “sprawling Chinese influence operation” is being employed largely for “intimidating Chinese dissidents abroad, including by impersonating U.S. immigration officials.” Though the exact number of users is unknown, the operation “appeared to involve hundreds of Chinese operators and thousands of fake online accounts on various social media platforms.”

OpenAI’s report “offers one of the most vivid examples yet of how authoritarian regimes can use AI tools to document their censorship efforts,” said CNN. In the case of China impersonating U.S. officials, this was done to “warn a U.S.-based Chinese dissident that their public statements had supposedly broken the law.” ChatGPT “served as a journal for the Chinese operative to keep track” of their own covert operations.

China is also geopolitically implementing ChatGPT, according to the report; several Chinese accounts “generated English-language emails to state-level U.S. officials or policy analysts working in business and finance, inviting targets to participate in paid consultations,” said Reuters. China’s use of AI also involves more targeted, nonpolitical scams: One group of ChatGPT accounts “used the chatbot to run a dating scam targeting Indonesian men and likely defrauded hundreds of victims a month.”

ChatGPT fights back

China’s AI use is heavy-handed, but there are signs that some of ChatGPT’s built-in safeguards are working. In October 2025, the chatbot “refused to assist an individual associated with Chinese law enforcement in planning an online campaign to discredit” Japanese politician Sanae Takaichi, said Bloomberg. The user allegedly asked ChatGPT to create a “plan that would amplify negative comments” about Takaichi, who became Japan’s prime minister later that month.

But ChatGPT “refused to provide advice on this plan,” said OpenAI’s report, and the user was forced to abandon their efforts. There is evidence, however, that the smear campaign against Takaichi “went ahead, likely using locally hosted Chinese AI models,” said Axios. And many of the other influence-peddling operations described in the report “reflect the same old tools and tactics that influence operators typically use in online campaigns — just supercharged with AI.”

OpenAI’s report “clearly demonstrates the way that China is actively employing AI tools to enhance information operations,” Michael Horowitz, a former Pentagon official and a professor at the University of Pennsylvania, said to CNN. The AI competition between China and the United States is “continuing to intensify,” and this is “not just taking place at the frontier but in how China’s government is planning and implementing the day-to-day of their surveillance and information apparatus.”

Is the world running out of RAM? Why is the Royal Artillery keeping looted treasure behind closed doors? And how can Heathrow’s car park cost £1.3 billion?

Olly Mann and The Week delve behind the headlines and debate what really matters from the past seven days.

A podcast for curious, open-minded people, The Week Unwrapped delivers fresh perspectives on politics, culture, technology and business. It makes for a lively, enlightening discussion, ranging from the serious to the offbeat. Previous topics have included whether solar engineering could refreeze the Arctic, why funerals are going out of fashion, and what kind of art you can use to pay your tax bill.

You can subscribe to The Week Unwrapped wherever you get your podcasts:

• Spotify
• Apple Podcasts
• Global Player

Artificial intelligence is increasingly entrenching itself in our society, and universities are no exception. From clubs and classes to professional events, AI companies have made themselves staples on campus, and it is not simply being used to cheat in class. Some experts worry that AI could influence higher education in new and potentially destructive ways.

How is AI being pushed in universities?

Google has worked with Purdue University, California Community Colleges and other schools to “offer AI courses, certificates and products such as its AI assistant Gemini,” while OpenAI has launched a “consortium with 15 leading research institutions,” and Microsoft is “offering eligible students 12-month subscriptions to its AI productivity tools at no cost,” said The Wall Street Journal.

Anthropic is also partnering with CodePath, the “nation’s largest provider of collegiate computer science education,” to “redesign its coding curriculum as AI reshapes the field of software development,” said the AI firm. The initiative will put its AI tools, such as Claude Code, “at the center of its courses and career programs,” said the company. The goal is to “enable students to learn to build with Anthropic tools” and “contribute to real-world open-source projects,” said the Journal. CodePath is incorporating the program into institutions that “cater to low-income and first-generation college students and include historically Black colleges and universities and Hispanic-serving institutions.”

Even before making its way into the curriculum, AI clubs started popping up on campuses all over the country. These are not just intended for computer science and STEM students. At least 16 law schools have founded AI clubs in the past two years as “future lawyers seek to understand the rapidly expanding technology and how it will affect their careers,” said Reuters.

What is in store for higher education?

AI’s growing presence in universities is no surprise. Companies across industries have “pushed their employees to adopt AI tools, and many are now asking in job interviews how prospective hires use the technology and whether they are willing to learn,” said the Journal. “Three-quarters of employers expect the new graduates they hire to have used AI tools, though most say colleges haven’t prepared them sufficiently.”

But as more students embrace the power of AI, educators have become increasingly worried about cheating, leading to an AI Cold War. Some students are “turning to a new group of generative AI tools called ‘humanizers,’” which “scan essays and suggest ways to alter text so they aren’t read as having been created by AI,” said NBC News. In response, “companies such as Turnitin and GPTZero have upgraded their AI detection software, aiming to catch writing that’s gone through a humanizer.”

Tweaking AI copy in an attempt to sound more human is a troubling sign that the “technopoly is thriving,” said Ronald Purser at Current Affairs. “Universities are being retrofitted as fulfillment centers of cognitive convenience. Students aren’t being taught to think more deeply but to prompt more effectively.” Meanwhile, AI companies seem to “look at college students as a strapped customer base to hook when they are most stressed,” said Matthew Connelly, a vice dean for artificial intelligence initiatives at Columbia University, at The New York Times.

India has long been a “centre for outsourced IT support” but, with the arrival of AI, there are rising concerns for the welfare of female workers in the industry.

As tech companies move to reap the benefits of using remote workers or employing people at lower cost in smaller towns and rural areas, more and more Indian women are finding work as data annotators, said the BBC. They help “fine-tune” the behaviour of AI models, said Business Insider, by labelling content as “helpful” and “natural-sounding” or flagging it as “wrong, rambling, robotic, or offensive”. Much of the content they must view is violent, abusive and disturbing.

'Psychological toll’

“Women form half or more of this workforce,” said The Guardian. Annotator roles are “promoted aggressively online”, promising “easy” or “zero-investment” job opportunities that are flexible and require minimal skills or training. In reality, annotators are exposed to about 800 videos a day, many containing pornography, sexual assault, child abuse and graphic violence.

“The world sees cleaner feeds” as a result but remains largely blind to the women who must absorb “the trauma” so the machines can learn what to block, said India Today. They are exposed to the “internet’s darkest material”.

Such exposure can lead to disrupted sleep, distorted social relationships and a protective “emotional numbness” that is “rarely acknowledged”. There is “limited mental health support”, even though “images linger long after shifts end”. Often working remotely, balancing other aspects of life, these women are left “unseen, unheard and exhausted”.

Their “psychological toll” is “intensified” by legal isolation, said The Guardian. They are bound by “strict non-disclosure agreements”, meaning they are often unable to speak to friends or family about the content they view at work. “Violating NDAs can lead to termination or legal action.”

‘Income without migration’

There’s an “estimated workforce of at least 200,000 annotators” in India’s rural towns and villages, according to US firm Scry AI, said Agence France-Presse. This amounts to “roughly half of the world’s data-labelling workforce”.

Women are seen by companies as “reliable, detail-oriented” hires, and “more likely to accept home-based or contract work”, said The Guardian. These jobs offer them “rare access to income without migration”, and a rare opportunity for an “upward shift”.

The “appeal is understandable”, said India Today. Women can feel the “empowering” force of paid work without having to leave their communities. Even “modest pay can support families, fund education, or provide a degree of independence” which might otherwise be limited.