Stuart Russell, a distinguished professor of computer science at the University of California, Berkeley, and president of the International Association for Safe and Ethical Artificial Intelligence, made the case in a Guardian column published Wednesday that the world is facing “intolerable risks” from unrestrained development of unsafe AI systems. Russell argued that recent events at Anthropic — one of the most prominent AI companies — illustrate the danger and the need for a regulatory framework that would require minimum safety standards before systems can be built and released.
Russell pointed to two recent events he described as “consequential.” In early June, Anthropic posted an article describing early signs of recursive self-improvement (RSI), a process in which an AI system devises ways to increase its own intelligence, creating a feedback loop that could accelerate its capabilities beyond human control. According to Russell, Anthropic’s post suggested that “the world should ‘slow or temporarily pause frontier AI development.’”
On June 12, the White House issued an export control directive banning access to Anthropic’s new frontier models, Fable 5 and Mythos 5, for all foreign nationals — including many of the company’s own key researchers, Russell wrote. Anthropic responded by shutting the models down entirely.
“The CEOs are telling us: ‘We’re on track to create superhuman intelligence, which has a good chance of causing human extinction,’” Russell wrote, adding that by “good chance” he meant a probability roughly equivalent to one in six — the odds of dying in a game of Russian roulette with a loaded revolver. “Yet governments reply: ‘That’s wonderful! Can we offer you a subsidy? Fast-track your permits?’”
Russell traced the technical developments that led to the White House’s action. He wrote that a few months ago, Anthropic’s Claude Code became so capable that its leading researchers “no longer write any code at all; they just describe ideas and experiments to Claude and it does all the work.” That accelerated the improvement cycle, Russell said, to the point where the latest iteration — called Mythos 5 — showed the ability to conduct end-to-end cyberattacks with no human assistance. He argued that if such systems were released “without cast-iron guardrails, almost anyone in the world could attack any country’s critical infrastructure at will.”
Russell noted that the UK’s AI Safety Summit in 2023 was an exception to what he described as a global pattern of ignoring risks. He called for a licensing regime similar to those that govern nuclear power, aviation, buildings, elevators, hairdressers and sandwich makers — requiring a minimum safety standard before a system can be built and released.
One leading AI CEO told Russell he did not expect serious regulation until a “Chernobyl-scale disaster.” Russell wrote that such a disaster would likely result in AI companies being “shut down immediately and perhaps permanently.” He argued that the recent change in White House policy, which he described as a “rare attack of common sense” after a previously deregulatory stance, suggested that a smaller incident — “a Three Mile Island” rather than a full Chernobyl — might be enough to spur action.
“Unrestrained development of unsafe systems leads to intolerable risks,” Russell wrote. “Governments can respond now, before the risks materialize, or they can wait and clean up the mess (if they still exist, that is).”