Bulletin · Euro Weeks
In the summer of 1974, scientists who had found the ability to rearrange the essence of life hit the emergency stop. Recombinant DNA (combining genetic material from different organisms) was so new that they worried about world-changing consequences if an engineered organism escaped from a lab. A committee led by Stanford biochemist Paul Berg recommended that experiments be halted until scientists worked out how to conduct them safely.
The following February, around 140 scientists gathered at the Asilomar Conference Center in California and eventually thrashed out safeguards so that research could continue. The National Institutes of Health (NIH) turned those into formal guidelines in 1976 and made federal funding conditional on following them. Half a century later, a researcher at one of the world’s leading tech companies has reached a version of the same conclusion, but too late.
Jacob Coxon , 27, worked at OpenAI and then at Anthropic. On September 8 he resigned. “Neither company is acting responsibly,” he wrote.
They are, he said, racing straight to self-improving superintelligence and “gambling with our lives.” He asked the labs to coordinate and called for a temporary ban on making models more capable. It was the Berg letter redux. ... AI had its own Asilomar moment nearly a decade ago, and squandered it.
In 2017, just like in 1975, researchers gathered at Asilomar for the Beneficial AI conference, convened by the Future of Life Institute, and produced 23 principles intended to guide the development of advanced artificial intelligence. Principle No. 5 was called “Race Avoidance.” It was simple: “Teams developing AI systems should actively cooperate to avoid corner-cutting on safety standards.” AI did not fail to have an Asilomar moment. It had one, and everyone went racing anyway.
The Familiar Story Evan Hubinger, who leads alignment stress-testing at Anthropic, responded to Coxon by saying he was correct. Hubinger wrote that he personally assigns a probability greater than 10 percent to AI killing all humans within the next decade, and that Anthropic does not yet have a plan to solve alignment for superintelligence and is “not clearly on track to.” Political concern is moving just as fast. On September 3, Democratic Representative Greg Casar and independent Senator Bernie Sanders announced the Ban Artificial Superintelligence Act , which would prohibit superintelligent systems and pause advanced AI development until a federal regulator has written safety rules.
“The leaders of the major AI companies publicly acknowledge that they do not fully understand the technology and that it is escaping their control,” Sanders said. The coalition backing a ban is bipartisan, and it includes Geoffrey Hinton, Yoshua Bengio, Steve Wozniak, Richard Branson, Steve Bannon and Glenn Beck. The Trump administration has embraced the opposite approach.
An executive order signed on December 11, 2025, declared that the United States is “in a race with adversaries for supremacy” in AI and set federal policy as the pursuit of global AI dominance through a minimally burdensome national framework. Its immediate target was state regulation, and it created a Justice Department task force to challenge state AI laws. ... What’s Really Going On The scientists who gathered in California in 1975 faced the same problem.
An individual laboratory could gain an advantage by conducting an experiment that others considered too dangerous. But biology provided regulators with something useful that today's AI researchers lack: the real world. The NIH guidelines divided recombinant DNA experiments according to risk and required corresponding degrees of physical and biological containment, graded from P1 to P4.
That meant, sometimes literally, concrete things like sealed rooms and airlocks. Anthropic is now trying to work out what the AI equivalent of that door might be. Its Frontier Safety Roadmap describes a “moonshot” project that would prototype what key workflows would look like under extreme security practices.
Yet its May update also contains an alarming admission: its initial discussions left it “skeptical that isolated networks are a feasible path forward inside the next 1-2 years,” when powerful AI may be developed. Meanwhile, OpenAI’s GPT-6 Astra is the first model the company has broadly deployed to reach the “Critical” cybersecurity level under its Preparedness Framework. OpenAI says that, with appropriate tools and access, Astra can identify previously unknown security vulnerabilities and develop ways of exploiting many hardened systems without a human directing each step.
That’s one difference between 1976 and 2026. But it’s not the only one. By the time researchers returned to Asilomar in 2017 and warned one another against a race, they were too late, because the commercial race already existed.
The principles called for cooperation but there was no authority that could ensure it. Nine years later, the U.S. government officially describes AI development as a contest for global supremacy, while laboratories around the world, but predominantly in the U.S. and China, compete for researchers, capital, computing infrastructure and technological leadership. Recombinant DNA was regulated before it was commercialized.
AI was commercialized first. Which is why the labs have now started asking to be regulated . On September 9, the day after Coxon resigned, OpenAI called for mandatory national AI safety requirements, including capability-based federal rules, independent assessments and incident reporting.
Anthropic reached a similar conclusion in February, when the third version of its Responsible Scaling Policy separated what the company can achieve on its own from what requires industrywide action. More than 1,000 scientists at leading AI labs also signed a letter in July warning that there is a real risk of “capability development rapidly accelerating beyond our ability to understand or control” the resulting systems. It asked the U.S. government for help.
It was, again, voluntary. There is, therefore, a temptation to interpret the original Asilomar conference as evidence that scientists in the 1970s possessed a civic mindset their descendants lack. The original Asilomar itself was controversial, however.
Scientists disagreed about the risks, and critics objected that a relatively small scientific elite was making decisions with consequences for the broader public. Its advantage was that the rules were written while everyone still had little to lose by accepting them. Today, there is too much to lose to stop the race.
Source: Newsweek
