Anthropic CEO Warns U.N. Council AI Mismanagement Could Endanger Humanity
Dario Amodei urged global safeguards covering biological-weapons misuse, model testing, loss-of-control risks and incident notification during a Sept. 23 Security Council meeting.
Speaking to the U.N. Security Council on Sept. 23, Anthropic CEO Dario Amodei warned that inadequate oversight of increasingly capable AI systems could endanger humanity and called for coordinated government safeguards.
The meeting was held at 3 p.m. ET (19:00 UTC) under the title “Artificial intelligence and international security.” Amodei participated by video and urged international agreements banning the use of AI to develop biological weapons.
He also called for global testing standards covering misuse and potential loss-of-control risks. The proposed framework would include ways for countries to evaluate and verify one another’s safeguards and notify others about significant incidents.
Amodei’s position puts advanced AI safety in a diplomatic setting focused on threats that could cross national borders. His earlier public warnings have described powerful AI as potentially resembling a “country of geniuses in a datacenter” by 2026 or 2027, with risks spanning biological weapons, cyberattacks and autonomous systems.
Anthropic’s Responsible Scaling Policy uses capability-based safety thresholds. As a model’s capabilities and associated risks increase, the policy calls for additional testing, security protections and limits on deployment.
The company has described that policy as a prototype for government testing and oversight, rather than a substitute for regulation. The approach reflects a broader argument that managing advanced AI risks will require action across the industry and among governments, not only within individual companies.
The Security Council meeting followed the previously announced plan for OpenAI CEO Sam Altman and Amodei to brief the body on AI risks. TokenPost previously covered the planned Security Council briefing.
