Altman Outlines Three Paths for Governing Advanced AI Development

OpenAI CEO Sam Altman discussed three approaches to advanced AI with employees during the week of Oct. 12, outlining choices that would put oversight and development in different hands.

One option was an international body empowered to enforce safety standards. A second would place leading AI labs in a shared organization. The third imagined a single developer with enough economic power to deny rivals access to computing resources. The conversation did not set a timetable or establish any of the options as OpenAI policy. The first two centered on oversight and organizational design; the third tied competitive power to computing access.

Computing resources shape which developers can keep pace in building advanced systems. That question is explicit in the third scenario, where one company’s economic position could affect competitors’ access. The international body and shared-lab approaches instead address safety enforcement and how development is organized.

OpenAI’s Sept. 21, 2026, public proposal calls for international technical standards for testing advanced systems and reporting incidents. Governments would decide whether to put the standards into law. The standards would not themselves act as licenses or require a model to receive approval before release.

In 2023, Altman and two OpenAI co-authors proposed considering an international authority for the most capable AI systems. Its possible powers included inspections and audits, giving an international body a direct oversight role. The current public proposal centers on technical standards and national decisions about law.

The discussion also connects to recursive self-improvement, the process in which AI takes an increasing role in developing later AI systems. The fully autonomous form is not happening today. Recursive self-improvement could accelerate progress, making the question of how to organize safety work relevant to future AI development.

At the U.N. Security Council on Sept. 23, Altman said: “No one person or company or country should be able to use the most powerful AI models to impose their worldview on everyone else.”