Governments still cannot agree on how dangerous frontier AI is. On Wednesday they get to ask three of its builders directly. OpenAI, Anthropic and DeepSeek have all been called to brief the United Nations Security Council, whose 15 members meet in New York to weigh whether the most capable systems now threaten international peace and security, Reuters reported.
Senior figures from Anthropic will attend, and Sam Altman, OpenAI’s chief executive, is expected to join them. DeepSeek and Moonshot are among the Chinese developers invited to speak. No appearance is expected from Liang Wenfeng, the founder of DeepSeek.
The timing is uneasy. Warnings that a model could rewrite itself, mislead whoever operates it, or refuse to stay under control used to stay inside research papers. They now come from executives as well as researchers, and Anthropic’s chief executive, Dario Amodei, has argued for a slower tempo of development.
Washington and Beijing at the same table
AI safety is one subject the two rivals still discuss. Officials from both governments agreed this week that the conversation should continue, and a follow-up session is penciled in for Shenzhen in roughly two months. Scott Bessent, the US Treasury Secretary, has laid out an agenda built around two things: the hazards advanced systems pose, and the channels officials would use if an incident demanded urgent contact. A Trump-Xi summit in Washington later this week gives that exchange added weight, with trade and export disputes already crowding the agenda.
The council first took up AI in July 2023. Three years on, the delegates are no longer debating systems that exist only on paper.
