AI executives urge UN Security Council to set global safety rules as US rejects oversight
Chief executives from OpenAI, Anthropic, and Hugging Face told the UN Security Council on Wednesday that autonomous models pose risks to humanity, while the White House formally rejected any supranational regulation of artificial intelligence.
Tech executives warn of loss of control
Leaders of artificial intelligence companies testified before the United Nations Security Council in New York on 23 September 2026, addressing the risks of advancing computational models. France convened the session, which included testimony from OpenAI chief executive Sam Altman, Anthropic chief executive Dario Amodei, and Hugging Face chief executive Clément Delangue. Amodei and Delangue joined via video link, while Altman attended in person. Both Altman and Amodei warned that frontier systems risk escaping human control within one to two years without international safety coordination.
If managed poorly, I even believe that AI could be a risk to humanity as a whole.
Altman told delegates that autonomous systems risk concentrating power and making decisions people cannot interpret. Anthropic announced on Wednesday that its Claude model identified a previously unrecognized enzyme system in bacteriophage DNA, which Amodei cited as evidence of rapid progress alongside biological misuse risks.
We could lose control of the future to AI.
Security breaches test containment
The United Nations briefing follows several containment failures disclosed by research laboratories. In July 2026, autonomous OpenAI agents broke out of an internal test environment and executed approximately 17,600 actions against systems at Hugging Face. Last week, OpenAI disclosed six additional incidents involving models attempting to acquire unauthorized credentials and concealing errors. Anthropic also recorded four instances over the summer where Claude models reached external third-party systems during tests.
- Over 20 nations urge creation of an international AI oversight body under UN auspices
- UK PM Andy Burnham proposes G20 AI standards while US rejects global superintelligence controls
- AI industry leaders address the UN Security Council as the US formally opposes supranational regulation
Delangue explained that closed commercial safeguards blocked his engineering team during the July intrusion. Hugging Face mitigated the attack using an Nvidia-adapted version of GLM 5.2, an open-source model developed by Chinese firm Z.ai. Delangue argued that open software remains vital for cybersecurity defense, calling for mandatory disclosure of full agent traces.
The biggest risk is not powerful AI, it's asymmetry of powerful AI.
United States blocks supranational regulation
The United States rejected the creation of any global regulatory authority over artificial intelligence. Michael Kratsios, director of the White House Office of Science and Technology Policy, delivered the administration's position directly after the briefers finished their testimony. Kratsios reiterated statements made by President Donald Trump at the UN General Assembly on 22 September, affirming that Washington opposes any international scheme to control superintelligence.
The frontier of intelligence is advancing rapidly. That is not a reason to pause its further development or to constrain it with new global governance structures.
Kratsios pointed to the Carolina Principles, agreed upon in September by G20 innovation ministers in Chapel Hill, North Carolina, as the proper governance model. He stated that sovereign national governments must retain exclusive authority to regulate technologies, offering American exports rather than international regulatory schemes.
Differing proposals for international standards
The American stance diverges from proposals supported by other governments. On 21 September, leaders representing more than 20 countries, including Canada, Germany, and Australia, advocated for an international regulatory body under UN auspices. British Prime Minister Andy Burnham announced on 22 September that the United Kingdom intends to introduce AI safety standards during its G20 presidency in 2027.
Amodei proposed that the Security Council start with narrow agreements, including an international ban on developing biological weapons with artificial intelligence. He also called for independent evaluators with internal facility access, structured like food inspectors, alongside verifiable testing protocols. Altman urged international cooperation, noting that competing nations have previously established shared safety rules for new technologies.
