OpenAI CEO Sam Altman laid out three approaches to safe superintelligence for employees last week, including a scenario in which one developer accumulates enough economic power to withhold compute from rivals. The account comes from The Information, citing people familiar with the matter. The remarks surfaced as the Trump administration prepares its first framework for overseeing developers of superintelligence, which makes the timing relevant for companies building on frontier models.
Three models from agency oversight to single entity
The first approach would create an organization modeled after the International Atomic Energy Agency to enforce safety standards. OpenAI publicly called for such an institution back in May 2023. It would inspect systems, require audits, and monitor compute and energy use. Such a body would depend on international cooperation, especially with countries building frontier models, including China.
The second approach was labeled the "Demis Plan" after Demis Hassabis, then head of Google AI lab DeepMind, who made a proposal about ten years ago. Hassabis envisioned an AI organization modeled after CERN. In Altman version, the major labs would merge into a single entity or operate under its umbrella, and that entity would supply cooperating countries with models.
The third approach was called the "Eliezer Plan" after AI safety theorist Eliezer Yudkowsky. In this scenario, a developer gains so much economic power through self-improving AI that it withholds compute from everyone else and, in the most extreme case, shuts down rival computing capabilities entirely. Altman raised it only for completeness, according to the report, while one source read the remark as a jab at a competitor given Yudkowsky ties to Anthropic.
What Altman safety debate means for AI adopters
For business users of AI, the first two models point to a future with centralized audits, inspections, and accounting of compute and energy. If an agency-style body or a single supplier structure emerges, access to frontier models could come with stricter compliance checks and reporting duties. Large companies with legal and security teams would absorb those procedures faster, while smaller firms would feel the added cost in procurement and vendor review.
The limits of the discussion deserve attention. Altman reportedly complained that too few viable proposals exist for a "speed limit" on AI development, while OpenAI cannot afford to lose ground to rivals. Both OpenAI and Anthropic are reportedly working toward initial public offerings, with OpenAI planning to decide within the next year which path to take. That commercial pressure means safety statements alone do not guarantee slower releases or stable model behavior.
The marker to watch is the US oversight framework and the parallel Standards Authority for Frontier AI being built by OpenAI, Google, and Anthropic. The government has so far relied on voluntary pre-approval reviews, plus June export controls on Anthropic over jailbreak concerns around its Fable model. If the new framework introduces binding audits or compute reporting, enterprise contracts and deployment timelines will change first.
