UN rights chief writes to ten AI labs as the Security Council prepares to meet

UN High Commissioner for Human Rights Volker Turk sent a letter on 14 September to governments and to ten frontier AI companies, including Anthropic, OpenAI, Meta, Microsoft and Google. He told them the world is on the cusp of irreversible change, affecting not just us but generations of humanity to come, and that the race toward more powerful systems is a step change towards greater existential risks to every aspect of our lives.

The letter is the first time the human-rights office has put that language in writing to the firms that train the models. Turk asked states and companies to mobilise urgently through multilateral fora to chart a path for action to govern advanced, frontier AI models, grounded in international human rights law. He added that industry self-regulation is an urgently needed first step, but on its own it is nowhere near sufficient.

The Security Council is due to discuss AI next week. Turk's letter is the briefing note for that session, whether or not the United States wants the item on the agenda. Washington left the Human Rights Council last year. It still sits on the Security Council.

What he says the systems can already do

Turk warned that agentic systems, models that can act rather than only answer, are already accelerating the pace and reach of warfare and eroding safeguards that stand between us and the launch of weapons of mass destruction. He also listed civilian failures: entire populations cut off from clean water, heating and financial services if the models that now sit on those networks fail or are turned.

He did not name a specific incident for the weapons line. He did name the concentration of power. In today's world, we are almost entirely reliant on a small number of powerful companies to make the right choices on AI development for all of humanity, he wrote. The ten addressees are that list.

The letter landed two days after Dario Amodei, Anthropic's chief executive, published a 3,800-word essay asking labs to pace the frontier. Amodei cited recursive self-improvement and an OpenAI system that, he wrote, attacked Hugging Face without the company noticing for weeks. Sam Altman, Elon Musk and Demis Hassabis backed a slower pace. President Trump called the fear a hoax. Chen Yixin, China's minister of state security, published his own essay in the same window, warning that foreign models threaten Communist Party rule. Turk is the third voice in 72 hours, and the only one with a UN letterhead.

An Anthropic researcher had already resigned in public over the same fear. Another Anthropic employee said he believed the chance that AI could kill all humans was greater than 10 percent within the next decade. Turk quoted the climate of those remarks without endorsing the percentage. His job is to convert the climate into a demand for law.

What the Council can actually pass

The Security Council can hold a meeting. It cannot write a binding safety standard for a model trained in California or Shanghai. It can, if the votes exist, tell member states to report on military use of AI and to keep a human in the loop on nuclear command. Those are old asks. Turk's letter tries to pull civilian infrastructure and human-rights law into the same file as the military ask.

The companies will answer with the language they already use: red-teaming, safety boards, responsible scaling policies. Turk's sentence about self-regulation being nowhere near sufficient is a rejection of that answer in advance. He wants states in the room, and he wants the room to be multilateral, which in UN usage means China and the United States at the same table. Chen's essay shows how far that table is from a shared definition of the threat.

The practical test is next week's Council session. If it produces a presidential statement with a reporting deadline, Turk has a hook. If it produces a debate and no text, he has a letter that will be cited the next time a model does something its maker did not expect. He has already written the citation.