By Hamza Faisal
OpenAI CEO Sam Altman has told employees the company must decide within a year which of three possible futures for AI safety it will back, according to a report by The Information — a choice that lands just as the Trump administration drafts its first framework for overseeing superintelligent AI.
The three options, as described in the report: an international body modelled on the International Atomic Energy Agency that would monitor AI development and enforce safety standards — an approach Altman appeared to favour; a CERN-like shared research organisation, dubbed the "Demis plan" after Google DeepMind CEO Demis Hassabis, through which major AI companies would provide models to partner countries; and the "Eliezer plan", named after AI safety researcher Eliezer Yudkowsky, in which a single developer achieves self-improving AI and uses its economic power to choke rivals' access to computing chips.
Why the choice matters now
Altman reportedly included the Eliezer plan only for completeness and opposes it. Some OpenAI employees viewed the remark as a possible dig at Anthropic, given Yudkowsky's links to the company, The Information reported. OpenAI's leaders have supported an IAEA-style agency since 2023, when Altman, president Greg Brockman and then-chief scientist Ilya Sutskever proposed a body that could inspect AI systems, require audits and track computing and energy use.
The industry scrambles for guardrails
The discussion follows security incidents involving AI models from OpenAI, Anthropic, Google and Meta, and comes as OpenAI and Anthropic move toward public listings — intensifying competitive pressure. OpenAI, Google and Anthropic are separately working on a US-based group called the Standards Authority for Frontier AI, which they aim to launch by early 2027, though the companies differ on how new models should be tested.
Anthropic has also proposed draft legislation that would allow the US Commerce Department to halt models capable of causing large-scale harm. And on September 29, executives from Anthropic, Google, Meta, OpenAI, Nvidia and xAI signed a voluntary accord that Trump described as "morally binding" — although it carries no legal liability.
A one-year clock
Under the US government's current voluntary framework, developers are asked to provide access to certain AI models up to 30 days before release. Altman's reported one-year timeline puts pressure on OpenAI to settle its own position as the industry keeps pushing forward — with the central question no longer whether advanced AI should be governed, but who gets to write the rules.



