Altman and Amodei brief UN Security Council on AI risk

On Sept. 23, 2026, France convened the U.N. Security Council’s 10228th meeting on artificial intelligence and international security; OpenAI CEO Sam Altman briefed in the chamber and Anthropic CEO Dario Amodei by video — alongside Turing Award winner Yoshua Bengio and Hugging Face CEO Clément Delangue — on frontier-AI upside and loss-of-control risk (UN News framed OpenAI and Anthropic briefing amid a “real and imminent” runaway-AI threat; Bengio: a threat “none can contain alone”) (news.un.org/en/story/2026/09/1168414; transcripts.un.org/en/sc/10228). Altman’s as-delivered OpenAI primary restates human control, no training of models labs cannot align and monitor (incl. as recursive self-improvement accelerates), and complementary national/international standards for capabilities, safeguards, oversight, and incident reporting (openai.com/index/sam-altman-un-security-council-remarks/). Amodei named two lanes — misuse (e.g. AI-enabled biological weapons) and loss of control — pledged Anthropic will “slow down as much as necessary” for safe releases, and asked the Council for narrow bans first (incl. AI bioweapons), evaluation/verification so states can check commitments, and common testing standards plus notification for security-significant AI incidents; Anthropic company remarks page 404 at publication — Amodei lines from a UN transcript + UN News / AFP / Guardian (from an automated transcript, not the official UN record). Briefing ≠ resolution, treaty, or binding lab commitment.

Nvidia’s Isaac ROS 5.0 ships an AI coding-agent skill — migrate-node-to-rosidl-buffer — that audits a CUDA-accelerated ROS 2 node, plans a minimal interface-preserving refactor onto ROS 2 Lyrical’s rosidl::Buffer plus Nvidia’s contributed CUDA buffer backend, and verifies zero-copy GPU-resident payload transport when publisher and subscriber meet runtime requirements (same host/device/user; supported RMW e.g. Fast DDS / Zenoh); all Isaac ROS 5.0 nodes are updated to that backend, with CPU fallback when peers are incompatible, demoed on a Depth Anything 3 TensorRT depth node and deployable on Jetson AGX Thor (developer.nvidia.com blog Sep 22, 2026).

Earlier coverage

Sep 23 5pm: Altman at the UN Security Council; Amazon Seller Central Claude/Quick; OpenAI MentalHealthBench. Sep 23 morning: Abbott TCEQ data-center permit halt; Muse human-concierge rollback; Snorkel $350M Series E; Unit 42 Continuous Frontier AI Defense; BC OpenAI Tumbler Ridge suit. Sep 22 5pm: Opus 5.5; GPT-6 Sol+Luna.

— Lark Pemberton