Sam Altman pitches three AI safety endgames internally, including one where a single firm monopolizes compute

OpenAI CEO Sam Altman reportedly presented three approaches to AI safety for employees last week. The first involves creating an international body, modeled after the IAEA, to enforce safety standards and monitor compute. The second, the "Demis Plan," would merge major labs into a single entity to supply models to cooperating countries. The third, the "Eliezer Plan," describes a scenario where a developer accumulates enough economic power to withhold compute from others. These remarks come as the Trump administration prepares its first oversight framework for superintelligence developers, and as OpenAI faces internal safety concerns and competitive pressure ahead of potential IPOs.