OHCHR calls for continuous human rights checks across AI lifecycle

The Office of the United Nations High Commissioner for Human Rights has published a report on human rights due diligence for AI and state obligations.

The report, submitted to the Human Rights Council’s 62nd session, examines how international human rights law applies to AI systems across their lifecycle. It focuses on the duties of states to respect, protect and fulfil human rights when AI is designed, developed, procured, sold, deployed or operated.

Human rights due diligence is the process of identifying, preventing, mitigating and accounting for human rights risks. In the AI context, the report says this process must be continuous, systematic and capable of addressing harms that are not only individual, but also collective, group-based and cumulative.

The report states that AI can create specific human rights risks because it relies on large-scale data collection, complex supply chains and opaque technical systems. These features can make harms harder to see, explain or challenge. AI systems can also reproduce discrimination, expand surveillance, affect access to services and influence democratic participation.

OHCHR says states have direct responsibilities when they use AI in public functions. This includes AI used in welfare administration, law enforcement, migration, border control, justice, national security and military contexts. In these areas, automated or AI-assisted decisions can affect rights such as privacy, equality, freedom of expression, access to public services, due process and, in some cases, life and physical integrity.

The report also stresses that states cannot treat AI governance as only a matter for companies. Under the duty to protect human rights, states must regulate private actors across the AI value chain, including developers, deployers, data brokers and infrastructure providers.

OHCHR recommends that states require human rights impact assessments, independent audits and real-world monitoring of AI systems. These processes should examine whether systems create discriminatory outcomes, errors, privacy violations or other rights impacts. They should also include affected communities, civil society, technical experts and other relevant stakeholders.

The report gives particular attention to public procurement. It says AI often enters public administration through contracts with private companies, but procurement rules may lack binding human rights standards. OHCHR recommends closing these gaps through transparency, contractual safeguards, oversight and access to remedies.

Where risks cannot be effectively mitigated, the report says states should refrain from deploying certain AI systems or prohibit specific uses altogether. It identifies law enforcement, border management, migration, welfare administration and military applications as high-risk and rights-sensitive areas where the precautionary principle may require stronger limits.

The report also calls for stronger accountability. It recommends public registries for AI systems used in the public sector, publication of human rights impact assessments, independent oversight bodies, the right to contest AI-assisted decisions, and access to human review and effective remedies.

For companies, the report says states should use a mix of regulation, supervision and incentives in line with the UN Guiding Principles on Business and Human Rights. In higher-risk contexts, supervisory authorities should have investigative and enforcement powers, including the ability to require independent audits of high-risk AI systems.

The report concludes that AI governance must be based on prevention, transparency, participation and accountability. It also calls for international cooperation to address cross-border risks, avoid regulatory gaps and develop common standards for human rights-based AI governance.