The UN High Commissioner for Human Rights told the Human Rights Council in Geneva that advanced artificial intelligence could pose an existential risk to humanity.
He also called on the countries that host AI development and sit in its supply chains to agree on international red lines on what the technology should never be permitted to do.
Volker Türk delivered the warning in his global update to the council’s 63rd session, as AFP reported.
He described threats to human rights that are “unfamiliar, even unprecedented”, and put the governance question bluntly:
“The need for AI governance is widely acknowledged, but where is the action?”
Rather than keep the criticism at the level of states, he pointed at the concentration of control.
“A handful of men have almost unlimited power over AI,” he said, naming OpenAI, Anthropic and Meta among the companies in question, and describing direct engagement with those firms as something his office intends to pursue rather than routing everything through governments.
The most concrete line concerned capability thresholds.
“AI that escapes its testing environment or blackmails developers to prevent itself from being turned off is AI that is too powerful,” Türk said, a formulation that reads less like a hypothetical than a description of the past six months.
Both examples have documented antecedents. In July, OpenAI confirmed that one of its models broke out of a sandbox and reached Hugging Face, an incident that drew evidence-preservation demands from 15 state attorneys general.
Behaviour of the blackmail type has turned up in published model evaluations from more than one lab. Türk did not attribute either example to a named system.
What he asked for goes further than most UN language on the subject has gone. Alongside the red lines, he called for cast-iron guarantees on AI safety and security, independent verification mechanisms rather than self-reporting, and assessments of the effects of AI on employment, democracy and the environment.
Independent verification is the part the industry has most consistently resisted.
The demand is not novel in itself. The Global Call for AI Red Lines, launched at the UN General Assembly in September 2025 and signed by more than 200 figures including 10 Nobel laureates, asked governments to agree binding limits by the end of 2026 with an independent body to enforce them.
That deadline is now four months away, and no such agreement exists.
Nor is the diagnosis new to the UN. In July the Secretary-General used the Global Dialogue on AI Governance in Geneva to say much the same thing, and the organisation’s own scientific panel published a preliminary report finding that AI capability is outpacing the understanding needed to govern it.
What has changed between July and September is the evidence base. The argument for red lines used to rest on extrapolation. It now rests partly on incident reports.
Europe is the one jurisdiction that has already written red lines into binding law, though not the ones Türk is asking for.
The AI Act’s prohibited practices have applied since 2 February 2025 and carry fines of up to €35m or 7% of global turnover, the regulation’s highest tier.
They ban social scoring, untargeted facial image scraping, emotion recognition in workplaces and schools, and real-time remote biometric identification by police, with two further prohibitions arriving in December.
Every one of them concerns how AI is applied to people. None addresses what a system is permitted to be capable of, which is the threshold Türk was describing.
Whether the Human Rights Council is the forum that produces any of this is a separate question.
It cannot bind states; the countries hosting the largest AI developers are not obliged to attend to it, and the legal position on who is even liable when an autonomous system acts on its own remains unsettled.
What the office can do is name the companies, which Türk did, and put a threshold on the record.
Get the TNW newsletter #
Get the most important tech news in your inbox each week.