Altman calls for global AI rules while a handful of labs still hold the controls

On September 23, Sam Altman urged the UN Security Council to establish common standards for the most capable AI systems. OpenAI thereby moved the discussion from corporate principles to the harder question of who measures, verifies and enforces the rules.

OpenAI brought three principles to the Security Council

Altman spoke at the Council's 10228th meeting alongside representatives of Anthropic, Hugging Face and the scientific community. He organized his case around three principles: people should remain central to decisions, scientific and economic gains should benefit people, and AI should empower individuals.

He called for complementary national and international standards covering capability measurement, risk assessment and meaningful human oversight. He also proposed rapid incident reporting and secure channels connecting governments, critical infrastructure operators and technical experts.

Safety is becoming a question of legitimacy

The most consequential part of the speech concerned concentrated power. Altman said major decisions should not be made by San Francisco labs alone and that companies must not substitute for democratic processes.

For governments, this is an invitation to create a shared language for evals, oversight and evidence. For AI companies, it is also a commitment to accept rules whose outcomes they may not control. That second part will distinguish a genuine policy shift from sophisticated political insurance.

The speech supplied goals without an inspection regime

Altman offered no binding audit design, independent access mechanism or deadline for reporting incidents. He also left unresolved the tension between global standards and national authority.

OpenAI remains a developer that builds, evaluates and releases its own models. Shared terminology is useful, but without independent measurement every vendor can still grade its own exam.

Audits, incident reports and access to evidence will decide the outcome

The next meaningful signal will be a protocol rather than another speech. Watch whether governments align definitions for serious incidents, permit independent evals and specify what evidence a lab must provide before deployment.

Verifiable standards that work for both open and closed models could give the UN process practical weight. Without them, human control remains an appealing phrase with no switch attached.

Lilith's verdict

Altman opened the inspection log before the Security Council, but OpenAI still writes most of its entries. Cooperation begins when an independent auditor can see the same models and incidents.