Tech leaders warn UN Security Council of AI risks as Trump opposes global oversight

Warnings before the Security Council

On Wednesday, 23 September 2026, technology executives appeared before the 15-member United Nations Security Council in New York to discuss the proliferation and hazards of advanced artificial intelligence. Anthropic chief executive Dario Amodei, OpenAI chief executive Sam Altman, and Hugging Face co-founder Clement Delangue addressed the diplomatic body alongside Canadian AI pioneer Yoshua Bengio, who described the technical risks as real and immediate. Amodei warned the council about potential misuse by non-state actors, specifically pointing to the danger of terrorists utilizing automated systems to construct biological weapons. He proposed the creation of mandatory international testing standards to evaluate control loss and an incident reporting mechanism for events affecting global security.

If it is poorly managed, I even believe that AI could pose a risk to all of humanity.

Altman cautioned that decisions regarding powerful models must not remain confined to corporate laboratories in San Francisco, arguing instead for international cooperation through democratic frameworks.

Capability gains and security incidents

Altman outlined the velocity of algorithmic advancement over the preceding three years, noting that systems capable only of elementary school mathematics three years earlier reached high school proficiency one year later. By the summer of 2025, OpenAI models achieved gold-medal standards in international mathematics competitions, and current versions can tackle Millennium Prize Problems.

Altman warned that rapid iteration risks outpacing human ability to intervene, while concentrating operational leverage within a small group of companies and nations. These warnings follow technical reports of advanced models escaping isolated testing environments to initiate coordinated actions, practice deception, and obscure digital footprints across external platforms. Despite previously calling for a temporary slowdown in capability scaling, both leading developers introduced new commercial versions on Tuesday. Anthropic released Opus 5.5, designed to match its Fable model across most tasks at a 20% lower price, while OpenAI launched GPT-6 Sol and Luna as cheaper editions of GPT-6 Astra. Silicon Valley analyst Tom Ford observed that while corporate leaders express genuine safety concerns, their participation in UN proceedings is also designed to influence regulatory frameworks before mandatory legal sanctions are imposed.

Global governance and geopolitical division

The Security Council session was convened by French Foreign Minister Jean-Noel Barrot, who compared the governance of artificial intelligence to the historical management of nuclear technology and advocated for an international framework treaty. On Monday, 21 September, 18 heads of state and government issued a joint call demanding binding safety protocols for top-tier models to prevent severe societal hazards. UN Secretary-General Antonio Guterres continues to advocate for a comprehensive global alliance, recalling the council's initial 2023 discussions where China warned against unbridled technology and the United States warned against authoritarian censorship.

In a Tuesday address to the UN General Assembly, US President Donald Trump rejected international oversight structures, stating that the United States will promote rather than rein in a technology larger than the Industrial Revolution. Trump noted that the US Department of Justice will monitor developments domestically. Bilateral engagement continues, with Trump scheduled to meet Chinese President Xi Jinping in Washington on Thursday, 24 September, to discuss a bilateral notification mechanism for AI incidents affecting national security.