Volker Türk Says Voluntary AI Safety Pledges Are “Nowhere Near Sufficient”

The race to develop increasingly capable AI requires more than company-led commitments, according to UN High Commissioner for Human Rights Volker Türk. In an open letter, Türk called for urgent multilateral action grounded in international human-rights law, warning that voluntary self-regulation is “nowhere near sufficient.” 7

What Türk asked governments and AI companies to do

Türk’s appeal was aimed both at states and at companies developing frontier AI—systems at the leading edge of capability. His core request was for coordinated safeguards that do not leave a small number of companies to decide which risks the public must bear. 7

The measures emphasized in reporting on the letter include:

• Binding, rights-based governance. States should strengthen regulation rather than rely on voluntary corporate pledges, and cooperate internationally to prevent gaps between national approaches. 7
• Continuous human-rights due diligence. AI developers should identify and address actual and potential human-rights impacts throughout an AI system’s lifecycle, not treat assessment as a one-time compliance exercise. The UN’s AI guidance describes due diligence as a continuous process involving risk identification, action, tracking effectiveness, and communication.
• Independent technical assessment and verification. Türk called for safeguards that can be checked independently rather than accepted solely on companies’ own assurances. 7
• Urgent multilateral action. The letter frames frontier-AI governance as an international problem: the relevant rules and safety guarantees need coordinated development across countries and institutions. 79

Why the warning came now

Türk’s intervention arrived amid a highly visible dispute inside the AI sector over how quickly leading labs should advance toward more autonomous and potentially self-improving systems.

Jacob Coxon, a researcher who had worked at both OpenAI and Anthropic, resigned from Anthropic in September 2026. He said the companies were racing toward self-improving superintelligence while “gambling with our lives.” 68 Coxon also said some people building AI believed it could “kill us all” by the end of the decade. That is a warning attributed to Coxon and other concerned developers—not an established forecast. 6

At the same time, an employee-led initiative called Pacing the Frontier asked the US government to support an international effort to create the technical and governance tools needed to deliberately slow frontier AI development when it outpaces humanity’s ability to oversee it safely. The initiative’s site lists 1,386 signatories from frontier-AI companies. 15

This emerging support for “pacing” matters, but it does not resolve the accountability question. Türk’s position is that voluntary commitments, even when supported by industry employees or leaders, cannot substitute for enforceable public safeguards. 7

Existing harms and long-term risks belong in the same framework

Türk did not frame the issue only as a remote extinction scenario. He said voluntary self-regulation was insufficient both to address existing harms and to prevent advanced autonomous models from circumventing human safeguards. 7

The available reporting does not provide a definitive, item-by-item list of current harms specified in the letter. It would therefore be inaccurate to attribute particular categories of harm to Türk’s letter beyond that broad warning. What is clear is his proposed test for governance: human rights must remain central while AI is designed, deployed, assessed, and governed. 7

The gap Türk identified

There is a notable convergence between some AI-industry voices and the UN warning: both recognize that capability development may need to be deliberately paced and independently evaluated. 715

The unresolved gap is political. Industry-backed statements can encourage coordination, but they do not themselves establish common thresholds, compel disclosure, require independent assessment, or create remedies and accountability when harms occur. Türk’s call is for governments to close that gap through urgent, multilateral regulation under international human-rights law. 79

For policymakers and AI buyers, the practical implication is straightforward: safety claims should be accompanied by evidence of ongoing rights-risk assessment, external scrutiny, and clear accountability—not simply voluntary promises from the companies building the systems.