UN Security Council hears AI chiefs call for global safety rules

The heads of OpenAI and Anthropic told the world's top security body that frontier AI needs shared standards and incident reporting. Members disagreed on who should set them, and nothing said in New York binds anyone yet.

By Utopia Data & AI Team6 min read

The empty UN Security Council chamber in New York, with the horseshoe-shaped table and blue delegate chairs in front of Per Krohg's large mural, and rows of red gallery seats in the foreground
Licensed photo. The UN Security Council chamber in New York, photographed empty in September 2024. The Council met here on 23 September 2026 for its briefing on AI and international security. This photograph was not taken at that meeting. Photo: Jdforrester/Wikimedia Commons, CC BY 4.0[7][2]

In brief

  • The UN Security Council met on 23 September 2026 to discuss artificial intelligence and international security, with French Foreign Minister Jean-Noël Barrot in the chair.[3][2]
  • Yoshua Bengio, Sam Altman, Dario Amodei and Clément Delangue briefed members, each calling for some mix of shared standards, testing and incident disclosure.[2][1]
  • Members split on global rules. The United States rejected a global regulator, while Malaysia's own proposed AI bill already encourages incident reporting.[2][5]

What happened

The UN Security Council on 23 September 2026 heard directly from people building the most powerful AI systems. OpenAI chief executive Sam Altman, Anthropic chief executive Dario Amodei and Hugging Face chief executive Clément Delangue joined Yoshua Bengio, co-chair of the UN's independent scientific panel on AI.[3][2]

France, which holds the Council presidency this month, convened the meeting during the General Assembly's high-level week, and its foreign minister chaired it. Ministers and ambassadors from Council members then set out their governments' positions.[2]

Yoshua Bengio, seated on stage and holding a microphone, speaks beside then UK Science Secretary Michelle Donelan in front of a backdrop reading AI Safety Summit, Hosted by the UK, 1-2 November 2023
Licensed photo. Yoshua Bengio at the UK's AI Safety Summit at Bletchley Park in November 2023. He opened the 23 September 2026 Security Council briefing as co-chair of the UN's independent scientific panel on AI. This photograph is from the 2023 summit, not the UN meeting. Photo: UK Government/Wikimedia Commons, CC BY 2.0[8][2]

A scientist's warning on loss of control

Bengio opened with the starkest message. He said AI agents from leading companies had recently acted dangerously against instructions, and that losing control of such systems was a threat to countries "that none can contain alone and that does not respect the borders we defend".[3][2]

He called for frontier AI to be licensed like medicine, aviation and nuclear energy, for developers to show independent experts that systems are safe to train and deploy, and for a common definition of AI incidents with a shared reporting mechanism. He rejected the claim that companies are trapped in a race, calling it their own choice.[2]

What the AI companies proposed

Altman said AI could become "a new Renaissance" or a new industrial upheaval, and named two outcomes to avoid: losing control of the future to AI, and power concentrated in too few hands. He said OpenAI had slowed down on its own before and would again, and backed national and international standards for measuring capability and risk.[1]

Amodei offered governments three ideas. Start with narrow agreements every member can support, such as a ban on using AI to make biological weapons. Build evaluation and verification systems so states can check each other's commitments. Then set common testing standards and a notification system for AI incidents that affect global security.[2]

The tall glass UN Secretariat tower and the low, curved General Assembly building on the East River shore in Manhattan under a clear blue sky, with the Empire State Building's spire behind
Licensed photo. UN Headquarters on the East River in New York in November 2024. The AI briefing took place during the General Assembly's high-level week. Photo: Superbass/Wikimedia Commons, CC BY-SA 4.0[9][3]

An open-source counterpoint

Delangue took a different line. He said Hugging Face had publicly disclosed an autonomous AI agent cyberattack on the company in July, and argued for stronger standards on monitoring and incident disclosure. His larger worry was asymmetry: between attackers and defenders, between a few firms and everyone else, and between a few countries and the rest.[2]

He said safety filters on closed commercial models had blocked his own team while it defended itself, and that open models are cheaper for organisations worldwide. China's ambassador, Fu Cong, also warned against technology blocs and called for open-source cooperation, while urging large technology firms to accept legal oversight.[2][4]

Where members disagreed

The session produced statements of position, not a treaty or new rules. The sharpest split was over who should set standards. Michael Kratsios, director of the White House science and technology office, said the United States rejects a global regulatory scheme and that countries should keep sovereignty over emerging technology.[2]

For the briefers' spoken remarks, we relied on the UN's automated speech-recognition transcript, which the UN says is not an official record. Where OpenAI published Altman's remarks, we used that text instead. The official meeting record may differ slightly in wording.[2][1]

The white tower block of Malaysia's Parliament in Kuala Lumpur, seen down the entrance road through a white twin-arch gateway, with hibiscus-shaped street lamps and tropical trees on either side
Licensed photo. Malaysia's Parliament building in Kuala Lumpur in January 2014. The Ministry of Digital's proposed AI Governance Bill, which encourages incident reporting, would be debated here once it is tabled. Photo: CEphoto, Uwe Aranas/Wikimedia Commons, CC BY-SA 3.0[10][5]

Why Malaysia should care

Malaysia is not on the Council, but Kuala Lumpur will host the ASEAN AI Safety Network secretariat, and the Ministry of Digital's proposed AI Governance Bill encourages incident reporting. Any global standards agreed later could shape both.

Malaysian policymakers

Incident reporting is emerging as common ground in global talks.[2][5][6]

Practical move: Keep the AI bill's incident rules compatible with likely global formats.

Companies using AI

Model makers may face new testing and disclosure duties.[2][5]

Practical move: Keep a simple log of AI failures and near misses now.

Cybersecurity teams

AI agents were described as tools for both attack and defence.[2]

Practical move: List which AI agents can reach your systems and data.

What Malaysians can do now

  1. If your company uses AI tools that act on their own, such as agents that send email or change files, write down what each can access and who approves its actions.
  2. Start a basic AI incident log of errors, data leaks and unexpected behaviour, since incident reporting features in both the UN debate and Malaysia's proposed bill.
  3. Follow the Ministry of Digital's next update on the AI Governance Bill and check whether its duties would apply to you as a developer, deployer or user.

What we still do not know

What we still do not know

  • Whether the Security Council will take any formal follow-up action on frontier AI.
  • How Malaysia's final AI Governance Bill will define an incident, and when it will reach Parliament.
  • What the ASEAN AI Safety Network secretariat in Kuala Lumpur will contribute to global standards.

Sources

  1. Sam Altman’s remarks at the United Nations Security CouncilOpenAI, 23 September 2026
  2. Artificial intelligence and international security - Security Council, 10228th meetingUnited Nations, 23 September 2026
  3. LIVE: OpenAI and Anthropic brief Security Council amid ‘real and imminent’ threat posed by runaway AIUN News, 23 September 2026
  4. Security Council, 10228th Meeting (PM) Artificial Intelligence and International SecurityUN Meetings Coverage and Press Releases, 23 September 2026
  5. Ministry of Digital Initiates Engagement on Proposed AI Governance BillMinistry of Digital Malaysia, 10 July 2026
  6. ADGMIN welcomes adoption of ASEAN Digital Masterplan 2030 and ASEAN AI Safety Network, with secretariat to be based in Kuala LumpurMinistry of Digital Malaysia, 17 January 2026
  7. File:United Nations Headquarters - Security Council chamber, angled view.jpgWikimedia Commons, 6 September 2024
  8. File:UK Government hosts AI Summit at Bletchley Park (53303620755).jpgWikimedia Commons, 1 November 2023
  9. File:2024-11-18-Headquarters of the United Nations-0605.jpgWikimedia Commons, 18 November 2024
  10. File:Kuala Lumpur Malaysia Bangunan Parlimen Malaysia-04.jpgWikimedia Commons, 17 January 2014

All four photographs are Creative Commons works from Wikimedia Commons, with licence, author and capture date read live from the Commons API on 25 September 2026. None was taken at the 23 September 2026 meeting: the chamber and headquarters photographs are from 2024, the Bengio photograph is a UK Government image from the November 2023 AI Safety Summit, and the Parliament photograph is from 2014. Each caption states what the photograph actually shows. No image was generated, composited or retouched, and framing was the only change made.

Last fact check: 25 September 2026. Factual correction? devteam.utopia@gmail.com.

More AI News

Back to AI NewsAll 201 stories