In the early 1970s biologists learned to splice genes from one organism into another. Recombinant DNA opened medicine and biology at once, and it frightened the people inventing it. They could not rule out that a modified organism might escape a lab and cause harm nobody knew how to contain. So they did something almost unheard of. They stopped.
In 1974 leading researchers, including future Nobel laureates, called publicly for a voluntary moratorium on the most hazardous recombinant DNA experiments. In February 1975 around 140 scientists, along with lawyers and journalists, met at the Asilomar conference centre in California to decide the terms under which the work could safely resume. They emerged with a framework of containment measures matched to risk, and research restarted under agreed safeguards. For advocates of pausing frontier AI, this is the reference point: a field that recognised its own danger and chose caution over speed.
What Asilomar genuinely got right
The achievement is real and worth stating clearly.
Scientists acted before a disaster rather than after one, which is the rarest and most valuable thing in the whole history of technology safety. They accepted restraint on their own work, including research that promised fame and progress, because the risk warranted it. And they built practical rules matched to the level of hazard, rather than either ignoring the danger or banning the field outright. It stands as proof that a research community can look at its own frontier, judge it too dangerous to rush, and hold back. That precedent is a direct rebuttal to the claim that pausing a powerful technology is naive or impossible, which is why it sits alongside the arms-control cases in our page on why acting is possible.
Why it is a harder model than it looks
Honesty about the precedent means naming why AI is a tougher case, because the differences are exactly the parts that made Asilomar work.
- It was a small, unified community. A few dozen labs, one scientific culture, people who knew each other. Frontier AI is a global contest among corporations and states worth trillions, with no shared table.
- The commercial stakes were modest in 1975. The biotech industry barely existed yet. AI arrives with the largest companies on earth and national security establishments already committed, which is the race dynamic Asilomar never faced.
- Restarting was the goal. Asilomar was designed to find safe terms to continue, not to stop indefinitely. Its containment model assumed you could bottle the hazard in a physical lab. A capable AI is not obviously containable in that way.
There is also a sobering epilogue. The Asilomar consensus held partly because it was a narrow moment with aligned incentives, and even then it depended on voluntary goodwill that a more commercial, more competitive field would strain. What worked among 140 collegial scientists does not straightforwardly scale to a fractured global industry under intense pressure to win.
Asilomar proves a pause is possible. It also shows the conditions a pause needs, and how few of them AI currently has.
The lesson the Foundation draws
Both halves of Asilomar matter. It refutes the fatalism that says nothing can be done, that a powerful technology cannot be slowed. Scientists did it, deliberately, and the sky did not fall. But it also shows that the informal, voluntary, collegial mechanism that worked in 1975 is not enough for AI, because AI lacks the small community, the low stakes, and the containable hazard that let goodwill suffice. The right conclusion is not that AI needs its own Asilomar. It is that AI needs what Asilomar could achieve, real restraint before the danger, delivered through the binding, verifiable, international machinery that a trillion-dollar global race requires. That is the argument of our plan.
Common questions.
The 1975 Asilomar Conference on Recombinant DNA was a meeting of around 140 scientists, along with lawyers and journalists, at a California conference centre to decide how to safely resume recombinant DNA research. It followed a 1974 voluntary moratorium in which leading biologists paused the most hazardous gene-splicing experiments because they could not rule out serious risks. The conference produced containment guidelines matched to the level of hazard, and research restarted under those agreed safeguards.
Because it is a clear case of a scientific community recognising that its own most powerful new technology might be too dangerous to rush, and voluntarily pausing until safety caught up. Advocates of pausing frontier AI point to it as proof that responsible restraint is possible, that researchers can hold back promising work when the risk warrants it, rather than an impractical or naive idea. It shows precaution before disaster, which is rare in the history of technology.
It acted before a catastrophe rather than after one, which is unusually far-sighted. Scientists accepted restraint on their own work despite the fame and progress it promised, because they judged the risk serious. And they produced practical rules matched to the level of hazard, avoiding both reckless continuation and an outright ban. Together these make it a genuine demonstration that a field can look at its own frontier, decide it is too dangerous to rush, and hold back.
Because the conditions that made Asilomar work are mostly absent for AI. Asilomar involved a small, unified scientific community with modest commercial stakes and a hazard that could plausibly be contained in a physical lab, and its goal was to find safe terms to continue. Frontier AI is a trillion-dollar global contest among corporations and states, driven by intense competition, with no shared table and a hazard that is not obviously containable. The informal, voluntary mechanism that sufficed in 1975 does not scale to that situation, which is why AI needs binding and verifiable international governance rather than goodwill alone.