AI’s architects made a choice. Now they’re calling it fate.

Recently Dean Ball, formerly the chief architect of Trump’s artificial intelligence action plan , issued a mea culpa for downplaying the risks of AI technologies that he knew could go “very, very poorly for human beings.” His essay was prompted by revelations of AI’s rogue capabilities, demonstrated by an alarming incident this summer in which an experimental application within OpenAI executed an unauthorized cyberattack on a company called Hugging Face.

Woven throughout Ball’s apology is an assertion that the emergence of dangerous, “self-sovereign” AI agents was inevitable. “There is probably nothing we could have ever done to avoid this outcome under even the best of circumstances,” he writes. Currently head of strategic futures at OpenAI, Ball acknowledges a culture of profound mistrust among Americans regarding pervasive digital technologies. He does not explore its roots, but laments that an atmosphere of suspicion makes a “genuine loss of control event” with AI harder to preempt.

Claims for inevitability regarding technological creations that pose existential threats have a long, colorful history. A close parallel with AI is nuclear weapons.

The initial rationale for the atomic bomb’s creation was to defeat Nazi Germany to the nuclear finish line, akin to the supposed imperative for the U.S. to outcompete China in the AI race. Once it was clear that Germany posed no nuclear threat, other rationales surfaced to guarantee forward momentum on a project whose irresistible allure lent it an air of inevitability. As physicist J. Robert Oppenheimer explained , “When you see something that is technically sweet you go ahead and do it and you argue about what to do about it only after you have had your technical success.”

OpenAI’s Sam Altman has likened himself to Oppenheimer — and AI to the Manhattan Project — citing the physicist’s belief in the inevitability of scientific progress. “Technology happens because it is possible,” Altman explained in an interview . He even mused that he and Oppenheimer share a birthday, as if to suggest their rendezvous with destiny.

Yet nothing in science or technology is destined to happen. The creation of potentially self-sovereign AI, like the drive to build atomic bombs, was a choice.

Now that the “ warning shot ” has occurred — OpenAI’s term for the Hugging Face incident — what can be done? Can we learn from history?

One suggestion, stemming from the extensive post-mortem conducted in the wake of this incident, is to create an independent agency akin to the National Transportation Safety Board, tasked with investigating plane crashes and making its findings available to the public. The parallel is imperfect: plane crashes are visible and widely documented. By contrast, the AI industry conducts itself in secrecy, ranging from nondisclosure agreements for data center creation to lack of transparency regarding data used to build models. Nevertheless, some such agency would be a start.

A second, more ambitious recommendation is an internationally agreed upon slowdown of AI development combined with minimal safety standards, a move called for by many in the AI industry. Given the U.S. economy’s dependence on the industry, this effort is likely to face resistance. But long-term benefits could greatly outweigh the risks.

Precedents exist for halting forward motion on powerful technologies. In 1975, international biologists, lawyers and journalists gathered at Asilomar in California to hash out a genetic truce in light of hazards posed by recombinant DNA . Rather than ban genetic research, the agreement called for strict, risk-based safety guidelines.

The research pause was brief but effective. As Nobel Laureate Paul Berg observed , Asilomar “marked the beginning of an exceptional era for science and for the public discussion of science policy.” To be sure, commercial and military interests were already lurking behind the scientific debate. Nevertheless, the conference allowed a contentious technology to flourish, with numerous public health benefits .

In 2017 an “ Asilomar for AI ” gathering was held on the grounds of the 1975 conference. There, a set of 23 guidelines was created, emphasizing “safe, ethical, and transparent AI” and shared prosperity. The principles stated that “development of advanced AI should be pursued only when its effects can be aligned with the long-term well-being of humanity.”

Judging by recent events, many of those principles appear to have been abandoned. The industry has helped to foster the culture of mistrust its chief architects cite as an obstacle to preventing greater AI harms. It remains a legitimate question whether voluntary self-regulation is possible or sufficient.

The European Union, meanwhile, has launched the AI Act , the first comprehensive and binding legal framework classifying AI systems by risk and setting obligations for how organizations develop, deploy, document and govern AI.

Ball’s widely read essay is simultaneously an apology for complicity and a denial that culpability exists where a technology is simply bound to emerge. Yet legislation like the EU AI Act is proof that the future of AI — and any technology — is a matter of choice, not fate. We should act on that knowledge now.

Lisa H. Sideris is a recent Public Voices fellow of the OpEd Project and a scholar of science, religion and environmental ethics at the University of California, Santa Barbara.