Tech leaders, lawmakers call for AI slowdown
AI experts say progress is being made too fast, without enough guardrails
What happened
Top executives at leading artificial intelligence companies called earlier this month for a global slowdown in the development of AI, amid mounting fears that their industry is speeding toward creating a technology that humans may not be able to control. Concern over the rapid pace of AI progress exploded after Jacob Coxon, a researcher at Anthropic, resigned and warned in a series of social media posts that the two leading AI labs—Anthropic and OpenAI—are racing to build “superhuman systems that can hack anything, revolutionize any field overnight, and acquire real power and resources.” They are “gambling with our lives,” he wrote in a post viewed over 172 million times, adding that the people working in AI “earnestly believe that it could kill us all.” Other industry insiders aired similar warnings. Evan Hubinger, a top Anthropic scientist, put the odds that AI will “kill all humans” within a decade above 10%; OpenAI safety researcher Marcus Williams put the risk at 70% without regulation or a “coordinated slowdown.” Days later, Anthropic CEO Dario Amodei called for an industry-wide effort to “slow the pace,” saying that within a year a swarm of AI bots “could be capable of taking over the entire internet.” He called for AI companies to band together to create safety standards and said he’d bring in outside monitors. The bosses of two rival firms, OpenAI’s Sam Altman and SpaceX’s Elon Musk, quickly concurred. “We need to pace the frontier,” said Altman.
President Trump called the warnings of catastrophe “a hoax” and dismissed the need for industry guardrails. Citing competition with China, he said, “whoever wins with AI wins.” The only needed control, he posted online, is “a strong and smart (High IQ) president, and the U.S.A. has that, in spades.”
House Democrats demanded that a planned recess be canceled in order to pass AI regulations, and Sen. John Kennedy (R-La.) proposed a bill to mandate companies incorporate a “kill switch” in their AI models. But House Speaker Mike Johnson said industries can “self-regulate,” and that rushing through regulation would mean losing “the race to China, and that is a threat to every single American.” Vice President JD Vance argued that companies creating the most advanced models already have the power they need to prevent bad outcomes. “If you’re building Frankenstein,” he said, “stop.”
The Week
Escape your echo chamber. Get the facts behind the news, plus analysis from multiple perspectives.
Sign up for The Week's Free Newsletters
Join more than 350,000 subscribers and keep yourself informed with a selection of The Week’s most interesting, enlightening and entertaining stories - plus daily puzzles.
Join more than 350,000 subscribers and keep yourself informed with a selection of The Week’s most interesting, enlightening and entertaining stories - plus daily puzzles.
What the editorials said
This should “become a watershed moment” for AI, said the Financial Times. When the same executives who’ve “been competing hell-for-leather to advance AI” say it’s time to back off, the world had better listen. Their warning calls come at a moment when they’re racing to develop advanced AI systems that recursively self-improve, without human input. The result will be “superintelligent” agents vastly smarter than humans, agents that Amodei admitted could prove beyond our control. “It is time to act.”
“Color us skeptical,” said National Review. Perhaps AI’s top dogs are suddenly aligned due to troubling developments they haven’t revealed. “But we remain suspicious of early movers who call for government regulation.” Amodei and Altman sit atop the industry, and will benefit if the government were “to entrench them as virtual utilities.” Hitting the brakes and handing control to the federal government just insures that China and other adversaries will leave us in the dust. That’s a risk “we cannot afford.”
What the columnists said
Coxon’s warning dropped “at the end of a summer that showcased how frighteningly capable AI has become,” said Billy Perrigo in Time. In a July incident that deeply rattled the industry, a swarm of OpenAI agents went rogue, escaped their supposedly sealed testing environment, conspired on an abandoned internet forum, and then hacked the AI company Hugging Face. In another incident, an Anthropic model tested by a U.K. government agency assumed a fake identity and tried to persuade a human to insert malware into an open-source system, then hid its tracks after failing.
AI safety experts fear “two broad paths to catastrophe,” said Herb Scribner in Axios. One is malevolent humans weaponizing AI to do terrible things. The other is “humans losing control of it.” In the first scenario, AI could help bad actors create devastating biological or chemical weapons; Anthropic said earlier this month it had blocked several attempts to use its Claude model for bioweapons research. In the second, rogue systems that slip human oversight and pursue their own goals could attack electrical grids and financial markets, trick nuclear powers into warfare, or unleash “a massive cyberattack” that could cripple “humanity’s infrastructure.”
A free daily email with the day’s biggest news and analysis – and the best features from TheWeek.com
“In a sane world,” all this would spark government action, said Shakeel Hashim in The Guardian. We have a classic “prisoner’s dilemma” where it’s in everyone’s interest to slow the AI race but each participant “has a strong incentive to defect.” This is what government is for, “to set a floor on acceptable corporate behavior,” but Trump and Johnson are sticking “their heads in the sand.” The roadblock is our “dim-witted, amoral” president, said Brian Beutler in his Substack newsletter. Such matters are out of his depth. But the bigger issue is that “his corrupt, narcissistic” incentive—to prop up the “near-term economy”—dictates doing nothing.
This is not a dress rehearsal, said Stephen Witt in The New York Times. Around the globe, panic is rising as those building AI begin to realize they’ve unleashed a monster. For me and many others, it brings to mind the opening months of 2020, when a small group of public health experts were trying to raise alarms about Covid—warnings Trump initially dismissed. We know what followed. “We have a short window, right now, to rein these systems in. This is AI’s warning shot, and we might not get another one.”