Altman outlined three AI safety governance options

The Information reported that Sam Altman told OpenAI employees last week about three approaches to governing advanced AI. The report said the comments came as the Trump administration was preparing its first framework for superintelligence developers.

The first option would have created a body modelled on the International Atomic Energy Agency. OpenAI had publicly called for such an institution in May 2023, and the article said it would inspect systems, require audits, and monitor compute and energy use.

The second option was described as the “Demis Plan”, after a proposal Demis Hassabis made about ten years ago. In Altman’s version, the major labs would merge into one organisation or work under its umbrella, and that body would provide models to cooperating countries.

The third option was named the “Eliezer Plan”, after Eliezer Yudkowsky. In that scenario, a developer would gain enough economic power through self-improving AI to withhold compute from everyone else and, in the most extreme case, cut off their computing capabilities; one source said the remark sounded like a jab at a competitor.

The article said Altman had complained that there were too few workable proposals for a speed limit on AI development. It also said OpenAI planned to choose a route within the next year, that OpenAI and Anthropic were reportedly working towards initial public offerings, and that OpenAI, Google and Anthropic were building the Standards Authority for Frontier AI while Meta and Nvidia pushed back on the idea.