United Nations

The United Nations (UN) is an international organization founded in 1945 and comprising 193 member states. Its Charter provides the institutional foundation for cooperation on international problems. Its AI-governance work includes expert reports, scientific assessment and international dialogue, with an emphasis on understanding risks and widening participation in decisions about AI.

Intergovernmental organization · International AI governance · Safety and human rights

Institutional role

The UN itself and the wider UN system are not interchangeable: the system includes programs, funds and specialized agencies with their own leadership, budgets and responsibilities. This profile covers central UN AI initiatives alongside selected work of UNESCO, the International Telecommunication Union (ITU) and UNIDIR, identifying the responsible institution for each contribution.

The Independent International Scientific Panel on AI and the Global Dialogue on AI Governance have different functions. The panel supplies scientific analysis; the dialogue provides a forum for governments and other participants. Neither is an AI regulator or enforcement body, and the panel does not prescribe policies.

Governing AI for Humanity: 2023–2024 reports

The Secretary-General's High-level Advisory Body on AI published an interim report in December 2023 and its final report, Governing AI for Humanity, in September 2024. The final report offered seven recommendations to address gaps in international AI governance. It advocated cooperation grounded in human rights and development, with institutions designed to complement existing initiatives and adapt as AI changes.

Its preparation included expert discussions, regional consultations and written submissions. The advisory body also commissioned an AI Risk Global Pulse Check and an AI Opportunity Scan to gather assessments of emerging developments. These activities placed risk identification alongside consideration of who can participate in AI governance and share its benefits. The temporary advisory body proposed a scientific panel and a global dialogue; its policy advice should be distinguished from the standing scientific mechanism subsequently established by the General Assembly.

Scientific assessment and global dialogue

Building on the 2024 Global Digital Compact, the General Assembly established the panel and dialogue through resolution A/RES/79/325 on August 26, 2025. Forty panel members were appointed in February 2026. Its remit covers AI opportunities, risks and impacts in the non-military domain. It produces annual assessments and may issue thematic briefs.

The panel released its Preliminary Report: Evidence-based assessment of opportunities, risks and impacts of AI on July 1, 2026. The first annual Global Dialogue took place in Geneva on July 6–7, informed by that assessment. The report's official summary warns that safeguards are failing to keep up with expanding capabilities. It also identifies a timing problem for policymakers: waiting for conclusive evidence can delay intervention until opportunities to prevent harm have narrowed.

The assessment's seven domains span technical advances, societal applications, economic effects, security and environmental consequences, human rights and democracy, individual autonomy and child safety, and governance and reliability. This scope treats AI safety as both a technical and societal concern. The panel describes the report as an initial assessment in an ongoing process; its next annual report is intended to inform the May 2027 dialogue in New York.

September 21, 2026: agent-control assessment

The Independent International Scientific Panel on AI released an advance unedited thematic brief examining the OpenAI–Hugging Face incident. It frames uncertain, potentially irreversible loss-of-control harm as a precautionary-principle problem. The assessment belongs to the independent panel; members serve personally, and its disclaimer distinguishes the report from official UN or government positions.

The brief surveys liability, insurance, incident reporting, independently reviewed safety cases and intervention controls as possible risk-management approaches. It discusses safeguards across borders and organizations. See September 21 digest coverage and United States policy on catastrophic AI risk.

UNESCO: ethics and prevention of harm

UNESCO's member states adopted its Recommendation on the Ethics of Artificial Intelligence in November 2021. It places human rights and dignity at the center of AI governance and connects safety with proportionality, privacy, fairness and human oversight.

Its safety principle addresses both unintended harm and vulnerability to attack. Its accountability provisions call for traceability, auditability, impact assessment and oversight, while retaining ultimate human responsibility. These are recommended governance practices; citing the framework does not establish that a particular system is safe or legally compliant.

ITU: standards and practical safety work

ITU's AI Standards for Global Impact: From Governance to Action (2025) records work from the AI for Good Global Summit and its International AI Standards Exchange, organized with ISO and IEC. Topics include trustworthy AI testing, cybersecurity and multimedia authenticity.

The report documents a standards database and collaboration on detecting deepfakes and verifying media provenance. It describes standards as voluntary tools that can support testing, risk management and regulatory compliance. Workshop findings are not themselves binding rules; the publication also cautions that its findings do not necessarily represent ITU or its membership.

Secretary-General's safety advocacy

On September 16, 2026, Secretary-General António Guterres called for national safeguards combined with international coordination. He warned that isolated voluntary pauses would be insufficient and urged protections supporting safety, transparency and accountability. These were policy recommendations, not an adopted international agreement.

September 22, 2026: General Assembly address

Guterres called for multilateral AI risk management with independent oversight, protection of children and human control over life-and-death decisions. See September 22 digest coverage.

September 23, 2026: Security Council debate

OpenAI, Anthropic and Hugging Face executives called for international safeguards for frontier AI at a Security Council meeting. OpenAI chief executive Sam Altman urged complementary national and international standards for risk assessment, incident reporting and human oversight. UK Foreign Secretary Ed Miliband urged common safety standards and government visibility into advanced models, while U.S. science adviser Michael Kratsios opposed a global regulator and emphasized national governance. See September 24 digest coverage.

Civil-society proposals and commentary

On September 22, 2026, policy director Mark Brakel published a FLI briefing proposing international reporting of serious AI incidents. It calls for independent evaluation, whistleblower protection, pre-release testing and national authority to halt deployments. The briefing recommends a common reporting mechanism, an expert group and cross-border communication channels, alongside limits on recursive self-improvement.

• Jane Munga’s September 22 commentary argues for greater attention to undersea-cable resilience in AI governance.

Related articles and coverage

• Organizations in AI Law and Policy
• International AI Frameworks
• Policy
• Research ICT Africa
• September 16, 2026: Guterres calls for coordinated international AI safeguards

UNIDIR: military-AI industry responsibility and dialogue

The United Nations Institute for Disarmament Research (UNIDIR) is an autonomous institution within the UN that conducts independent research on disarmament, arms control and international security. Its military-AI work is distinct from the scientific panel’s non-military remit.

In February 2026, UNIDIR launched a multi-year programme with the Office of the UN High Commissioner for Human Rights to develop a Framework of Responsible Industry Behaviour for AI in the Military Domain. It seeks a common baseline for industry conduct, procurement and public-private partnerships. Its intended principles and practical guidance are voluntary; participation in their development does not imply endorsement or binding commitments.

UNIDIR’s Centre of Excellence on AI, Peace and Security opened in June 2026. The listing for its October 22 New York event announced plans to launch a multistakeholder Forum and expressions of interest for organizational participation. It describes intended consultation and collaboration, not a new regulator or binding treaty; the listing alone does not verify the event’s outcomes. See October 20 news coverage.