The rapid development of artificial intelligence has reached a point where the consequences of decisions being made today could extend far beyond the present generation, UN High Commissioner for Human Rights Volker Türk warned on Monday, calling on governments and leading AI companies to accelerate efforts to govern increasingly powerful systems.
In a letter addressed to States and AI developers, Türk described the world as being at a “watershed moment”, as companies and researchers advance AI at what he called a “dizzying speed”. Investment in AI and its underlying infrastructure is breaking records, while billions of people are already using the technology to support scientific discovery, make information available across languages and enhance agricultural production.
But the High Commissioner warned that the benefits of AI cannot be separated from the decisions governing its development and deployment. Whether the technology helps address the climate crisis, improves health and education, strengthens public services or enhances human well-being will depend, he said, on choices being made now.
Those choices have become more urgent because the harms associated with AI are no longer hypothetical. Türk pointed to documented discrimination driven by AI, the automation of essential services in ways that can leave already marginalised people further behind, and the growing use of data and surveillance technologies that threaten privacy on a potentially irreversible scale. AI-powered disinformation campaigns, he added, are already distorting public debate, weakening social cohesion and placing additional pressure on democratic institutions.
The emergence of increasingly capable AI agents raises the stakes further. Unlike systems limited to individual tasks, these agents can translate goals into action, conduct complex workflows and operate with limited or no human supervision. According to Türk, this creates risks on a scale that existing approaches to AI safety may not adequately address.
“Recent events have shown that advanced systems can produce behaviour outside their intended operational parameters, circumvent safeguards, practice deception, and single-handedly pursue objectives in ways that their developers did not anticipate,” he wrote.
The potential consequences, he warned, extend from individual rights to the functioning of societies themselves. Failures involving agentic AI systems could bring essential services to a standstill, fracture communications, destabilise critical infrastructure and threaten democratic institutions. Entire populations could potentially be cut off from clean water, heating and financial services. Türk also warned that agentic AI is already accelerating the pace and reach of warfare while eroding safeguards intended to prevent the launch of weapons of mass destruction.
“We are on the cusp of irreversible change, affecting not just us but generations of humanity to come,” he said.
The High Commissioner therefore argued that AI safety must be understood as more than the technical reliability of individual products. It must also encompass the protection of people, communities, institutions and future generations. Fundamental rights, including the rights to life, food, privacy, health and security, are potentially at stake.
Yet accountability remains a major unresolved issue. Türk noted that the world is increasingly dependent on a relatively small number of powerful companies to make decisions about AI development that affect humanity as a whole. Calls within the industry to slow development have intensified, including through the “Pacing the Frontier” letter, which has received support from employees, early leaders of AI and scientists. A proposal supported by the heads of three leading technology companies has also emerged.
While Türk acknowledged that some may view such calls as reflecting hype or self-interest, he argued that their breadth and detailed basis within the industry should raise serious concern.
The immediate challenge, he said, is to ensure that meaningful AI governance develops at a pace capable of matching the accelerating development of the technology. At the same time, governments and companies must not neglect existing AI systems and the need to strengthen regulations and accountability for harms that are already occurring.
The responsibility falls particularly heavily on the companies developing frontier AI, which Türk said are based mainly in the People’s Republic of China and the United States of America. He called on them to cooperate across the industry to understand and strengthen existing controls, including automated monitoring of model behaviour and the sharing of lessons from safety incidents.
He also called for robust human rights due diligence incorporating agreed safety standards, with continuous improvement and independent internal monitoring to strengthen transparency and accountability. As an immediate precaution, companies should also limit the use of recursive self-improvement in frontier AI development until research has established how it can be carried out safely.
For governments, Türk called for independent and technically competent verification of agentic capabilities, with regulators given access to models, documentation and testing environments. States should mandate continuing human rights due diligence, require companies to report serious incidents during testing and deployment, and empower an independent body to investigate such incidents immediately with full access.
Companies should also be required to maintain safety plans meeting agreed standards, backed by liability regimes where inadequate safeguards cause harm. Türk stressed that AI systems should not receive legal personality and that responsibility for the actions of AI agents must remain with identifiable human beings.
The High Commissioner also rejected approaches that isolate AI governance within individual countries. Competition between jurisdictions, he warned, could encourage a “race to the bottom”, making international alignment essential. Industry co-regulation could provide an urgently needed first step, but, he argued, it would not be sufficient on its own.
The proposed response is therefore international. Türk urged States and companies to work through multilateral forums to develop a framework for governing advanced frontier AI grounded in international human rights law. The Global Dialogue on AI Governance, which first convened in Geneva in July and is scheduled to meet again in May 2027, was identified as an important forum for translating evidence into policy, building consensus and facilitating international cooperation.
The stakes, Türk argued, extend beyond the immediate generation of developers, regulators and policymakers. Young people entering adulthood today will inherit the consequences of decisions over which they have had little influence.
“No country can govern this technology alone,” Türk wrote. “No company should be able to decide by itself which risks the world must accept.”
His warning ultimately places the central question of AI governance beyond technological progress itself: who decides how much risk society should accept, who remains accountable when powerful systems cause harm, and whether fundamental rights can keep pace with a technology developing faster than the rules designed to govern it.
