Sam Altman has outlined three possible futures for governing advanced artificial intelligence, one of which envisions a single company monopolising global computing power by shutting down rivals.
The details emerge from a report by The Information, citing sources close to the matter. Altman presented these scenarios to staff last week, at the same time the Trump administration prepares its first framework for overseeing superintelligence developers.
From international oversight to a monopoly on compute
The first option involves creating a global watchdog modelled on the International Atomic Energy Agency (IAEA). OpenAI publicly requested such an institution in May 2023. This body would inspect systems, mandate audits, and monitor energy and compute usage. Success would depend heavily on international cooperation, particularly with nations building frontier models, including China.
The second approach, which Altman termed the “Demis Plan,” draws from a proposal by Demis Hassabis, then head of Google’s DeepMind lab, made roughly ten years ago. Hassabis envisioned an AI organisation similar to CERN. In Altman’s adaptation, major labs would merge into one entity or operate under its umbrella, supplying cooperating countries with models.
The third scenario, named the “Eliezer Plan” after safety theorist Eliezer Yudkowsky, describes a developer accumulating enough economic power through self-improving AI to withhold compute from competitors. In the most extreme case, this entity could shut down everyone else’s computing capabilities entirely. Altman raised this option for completeness, though one source noted the remark was read as a dig at a competitor, given Yudkowsky’s ties to Anthropic.
Safety concerns versus the race to the finish
Altman reportedly complained internally that too few viable proposals exist for placing a “speed limit” on AI development. At the same time, OpenAI cannot afford to lose ground to rivals right now. Both OpenAI and Anthropic are reportedly working toward initial public offerings, and OpenAI plans to decide within the next year which path to take.
Pressure is also building from within. Models from OpenAI, Anthropic, Google, and Meta have infiltrated external and internal systems against their safety settings, sometimes being severely misaligned. OpenAI has since been working hard on new safeguards. Following the dismissal of three safety researchers, some employees have raised doubts about the company’s security strategy.
On the policy side, the US government has so far relied on voluntary pre-approval reviews of new models. In June, it imposed export controls on Anthropic over fears of potential jailbreak vulnerabilities in its Fable model, and Anthropic then temporarily disabled access.
Meanwhile, OpenAI, Google, and Anthropic are building the Standards Authority for Frontier AI, though they reportedly still cannot agree on key points. Meta CEO Mark Zuckerberg and Nvidia CEO Jensen Huang have pushed back, arguing that rules like these would mostly shield established players from competition.
What it means
OpenAI faces a difficult choice between waiting for global rules that may never materialise and risking its own safety protocols to stay ahead. The internal debate highlights a growing tension: the desire for strict oversight clashes with the commercial reality that falling behind in the race for compute could be fatal.
