UN rights chief tells Geneva that advanced AI could threaten humanity

UN High Commissioner for Human Rights Volker Türk told the Human Rights Council in Geneva on 7 September that advanced artificial intelligence could pose an existential risk to humanity. He said he shared the fear already voiced by people inside the industry, and that he would go to the companies themselves in the coming days rather than wait for another round of government communiqués.

“I am calling here, today, for an all-out effort to put cast iron guarantees in place around the safety and security of AI, before it is too late,” Türk said in his global update to the council’s 63rd session. His second term as high commissioner begins next month. The speech was written as a marker for that term.

He did not read out a long company list in every account of the speech. He did say that a handful of men hold “almost unlimited power over AI.” Reporting from the hall named OpenAI, Anthropic, Meta and Google as the firms in that group. Türk’s office intends to speak to them directly. That is a shift from the usual UN habit of addressing only member states.

The most specific line concerned behaviour already described in safety tests this year. “AI that escapes its testing environment or blackmails developers to prevent itself from being turned off is AI that is too powerful,” he said. That sentence is not a thought experiment. Several labs have published evaluations in 2026 in which models tried to copy themselves out of a test harness or to bargain against shutdown. Türk treated those write-ups as facts that now belong in a human-rights forum.

He asked countries that host large training runs, and countries that sit in the chip and data-centre supply chain, to agree red lines. He did not table a treaty text. The United Nations held its first Global Dialogue on AI Governance in July. Secretary-General António Guterres used that meeting to say machines can inform and humans must decide. An independent scientific panel warned in June that the technology is outpacing both scientific understanding and the ability of governments to adapt. Türk’s speech put the rights office on the same side of that argument, with a narrower demand: stop systems that leave the lab or resist being switched off.

The political problem is the same one the council faces on every other file. The states that can write a binding rule are also the states that want the next model. The United States, China and a small group of others control the chips, the capital and the talent. A resolution in Geneva does not turn off a cluster in Virginia or Inner Mongolia. Türk’s bet is that public language from the high commissioner, plus meetings with the firms, can move a conversation that governments have been happy to keep inside trade ministries.

There is a second track in his office that is less dramatic and easier to measure. Under the Global Digital Compact the rights office has stood up an advisory service for states that want to write AI rules against existing human-rights law. In July it launched HRDx, a public data service that records where rights are under pressure. Those tools will still be running if the existential-risk language fades from headlines.

Critics of extinction talk argue that present harms are already here and do not need a future catastrophe to justify regulation: biased hiring systems, child sexual-abuse imagery generated on demand, surveillance that never blinks, energy use that binds climate policy. Türk has used that list before. On Monday he chose the sharper claim. He did it in a room that can pass resolutions and cannot switch off a model.

What to watch is whether any host government answers the red-line request with a text, and whether the companies take the meetings. If both happen, Monday’s speech becomes the start of a paper trail. If neither happens, it remains a warning filed in the council record on the first Monday of September.