TL;DR
For the first time, the CEOs of the world's two most prominent AI labs addressed the UN Security Council directly, proposing binding international safeguards including compute thresholds, mandatory safety audits, and a ban on AI-assisted bioweapon development.
What happened
- On September 23, 2026, OpenAI CEO Sam Altman and Anthropic CEO Dario Amodei briefed the UN Security Council during the high-level week of the UN General Assembly in New York.
- Amodei proposed mandatory compute thresholds: any entity training models above a set computational scale would face third-party safety audits and hardware registration with an international authority.
- Amodei also called for a ban on AI-assisted biological weapon development and international evaluation mechanisms to verify compliance with agreed commitments.
- Altman warned that autonomous AI systems could outpace institutional response capacity and argued no single company or country should use powerful AI to impose its worldview on others.
- The session was reinforced by a concurrent report from UN human rights chief Volker Turk, who urged member states to regulate AI to avoid "existential risks."
Why it matters
- The briefing marks a strategic pivot by leading AI labs: rather than resisting oversight, they are now actively lobbying for structured regulation to set liability boundaries and prevent a safety race to the bottom.
- Both executives backed creation of an independent international AI science panel modeled on the IPCC, signaling appetite for a permanent multilateral governance body with real authority.
- Autonomous AI agents, capable of browsing the web, writing code, and executing financial transactions independently, were flagged as a novel and urgent security vector requiring treaty-level controls.
- The EU AI Act is currently the most advanced national framework, but geopolitical friction, including pushback from nations citing technological sovereignty, blocks any binding Security Council resolution.
- Unlike nuclear materials, AI is software-based and borderless, making enforcement dependent on unprecedented compute-tracking transparency that raises serious IP and national security objections.
What to watch next
- Whether the Global Digital Compact's proposed independent AI science panel advances from concept to institution, and which nations agree to seat and fund it.
- Whether compute threshold proposals gain traction in multilateral negotiations or get blocked by major AI-producing states outside the EU regulatory orbit.
- How quickly autonomous agent deployments by OpenAI and Anthropic themselves are reconciled with the safety commitments their CEOs just made before the Security Council.
Originally published on Present of AI, a daily source-linked AI news timeline. Read the full timeline or browse the open dataset.