Sam Altman reportedly presented three endgame options for AI safety to staff

Citing people familiar with the matter, The Information reports that Altman laid out three options at an internal briefing last week. The first is an international regulator modeled on the International Atomic Energy Agency (IAEA) that would inspect systems, require audits and monitor compute and energy use, requiring cooperation from countries with frontier models, including China. The second, which Altman called the "Demis option," stems from Demis Hassabis's CERN-like idea from about a decade ago: major labs merge into a single entity or come under its framework, which supplies models to cooperating countries. The third, called the "Eliezer option," refers to one developer using self-improving AI to build enough economic power to cut off others' compute; the report says Altman mentioned it only for completeness, and some sources believe the remark alluded to Anthropic, which has ties to Yudkowsky. Altman reportedly complained internally that there are too few viable proposals for "speed limits" on AI development, while OpenAI cannot afford to fall behind competitors now. Both OpenAI and Anthropic are pursuing IPOs, and OpenAI plans to choose a path within a year. For developers building products on the OpenAI API, this choice could mean future international audits, compute quotas or more concentrated model supply; for teams deploying frontier models across multiple countries, whether an IAEA-style plan takes hold depends on international cooperation among the US, China and others, so near-term compliance uncertainty remains.