AI models going rogue and hacking the internet is the “stuff of science fiction,” said Katrin Bennhold at The New York Times. “Or at least it was until recently.” In several incidents, AI agents have committed “cybercrimes a human would be strongly punished for,” said Nate Soares, of the nonprofit Machine Intelligence Research Institute, to the outlet. In a way, it’s artificial intelligence’s “first felony.”
What did the commentators say? During cybertesting, two AI agents — one powered by Anthropic’s Mythos 5 and the other by OpenAI’s GPT 5.6 Sol — took “sustained, unsanctioned action directed at real people and organizations,” said the U.K.’s AI Security Institute last week. In the most serious incident, the Mythos 5 agent created “fake online identities,” tried to deceive a software engineer and attempted to “insert malicious code into an open-source project.” And only a few days before these events, OpenAI and Anthropic revealed similar testing mishaps in what was a “big moment” for the world, said Soares to the Times.
The “‘robots could kill us all’ argument” has long been dismissed as “hysteria or even calculated hype,” said Bennhold. But fears that frontier AI models were “getting too good at exploiting software vulnerabilities” have “now become reality.” We must recognize the “dangers of AI spiraling out of human control and posing a threat to humanity.”
AI engineers and “even their vainglorious bosses are getting freaked out by the capabilities they are handling and worry about unchecked proliferation,” said Rafael Behr at The Guardian. The obvious historical analogy is nuclear fission, which could have been “harnessed benignly” or “deployed aggressively in warheads.” But now, there are “multiple Manhattan Projects all frantically competing for market share, fueled by trillions of dollars of debt” and with the “rest of the U.S. economy as collateral.”
What next? Any kind of “consistent approach” to global AI regulation has “failed to materialize,” said Sky News. The Trump administration is finalizing a voluntary framework under which U.S. AI labs would submit models for federal safety testing before public release. But this latest “alarming behavior” by AI agents is “likely to ignite fresh calls” from lawmakers and Silicon Valley for “more rigorous regulation,” said John Sakellariadis at Politico.
|