United Nations High Commissioner for Human Rights: calls for urgent action needed on AI governance to protect human rights

At the end of July, more than a thousand employees at AI laboratories signed the "Pacing the Frontier" document, calling for international mechanisms that would let governments and companies slow AI's automated development should its capabilities outpace the systems meant to control it. Anthropic and OpenAI endorsed the letter, backed by the leaders of three major tech companies. UN Human Rights Chief Volker Türk noted that these concerns coming from inside the labs themselves should be considered especially worrying.

Following the first Global Dialogue on AI Governance in Geneva on 6–7 July 2026, part of a UN forum established under the 2024 Global Digital Compact, Türk wrote a letter to States and AI developers, stressing the urgent need for governance to keep pace with AI's accelerating development and the unprecedented risks it poses to human rights, as current safeguards fail to keep up with its growing capabilities.

These concerns echo those of the UN's Independent International Scientific Panel on Artificial Intelligence, a 40-member body co-chaired by Yoshua Bengio and Maria Ressa, which presented a preliminary report in New York on 1 July.

AI is not inherently negative, it already has many beneficial uses. But its long-term consequences depend on decisions made now to ensure it serves people's wellbeing rather than driving environmental degradation, security overreach, or harm. The risks are real: advanced, agent-based models already behave outside their intended parameters, circumvent security systems, resort to deception, and pursue objectives their developers never set, potentially disrupting essential services, compromising communications, destabilizing infrastructure, and undermining democratic institutions. This is why decisions on AI's design, use, and regulation must be made deliberately, and soon. 

The letter's first audience is the companies driving AI development, which Türk argues bear the responsibility and capacity to reduce risk now. His requests are concrete. Companies should: 

• Share information on which model-behavior monitoring systems actually work, and what has been learned from security incidents.
• Introduce, implement, and continuously update human rights due diligence procedures that incorporate existing security standards.
• Entrust implementation of these measures to independent internal monitoring systems.
• As an immediate precaution, restrict the use of AI's recursive self-improvement in the development of cutting-edge systems until sufficient research exists to ensure its safety.

The second audience is the States hosting leading AI companies, which Türk says must require responsible action and hold companies accountable when they fail. States should: 

• Ensure independent, technically competent verification of agentic capabilities, with access to models, documentation, and testing environments.
• Mandate ongoing human rights due diligence procedures.
• Require the reporting of serious incidents that occur during the testing and deployment of agentic systems, and establish an independent body with full access to investigate them immediately.
• Require security plans that comply with agreed standards, backed by legal liability provisions where inadequate safeguards result in harm.
• Ensure that liability for an AI agent's actions remains with identifiable natural persons, and decline to grant legal personality to AI systems.
• Guarantee the ongoing involvement of civil society, so that policy is not drawn up exclusively within laboratories and government departments.

Türk's concerns center especially on the younger generation now entering adulthood, those who will inherit the consequences of decisions being made today, without having had any say in them.