On 21 March 2024 the 78th session of the UN General Assembly adopted resolution A/RES/78/265 without a vote. The text was proposed by the United States and co-sponsored by more than 120 member states. It calls for seizing the opportunities of safe, secure and trustworthy AI systems to advance sustainable development, and stresses human rights, privacy, transparency and accountability. It is the first stand-alone UN resolution on AI, but its weight has to be measured against the character of soft law.
What unanimous adoption means
No vote means no objections, and it also means the text had to accommodate positions that are far apart. More than 120 co-sponsors span jurisdictions with very different regulatory instincts; what they could agree on is a minimum. “Safe, secure and trustworthy” therefore works as shared vocabulary rather than an executable standard. The resolution defines nothing, prescribes no assessment method and creates no supervisory mechanism.
That does not make it insignificant. In international governance, consensus is infrastructure: it gives later treaty negotiations, bilateral dialogues and domestic legislation a reference point. A company that judges an international document only by whether it is enforceable will miss its role as a signal of direction.
Where human rights, privacy and accountability land
The resolution places human rights, privacy, transparency and accountability side by side. All four recur in EU, UK and US regulatory texts, but in the UN setting they describe obligations of states more than of companies. The transmission runs through soft law into domestic law, and through domestic law onto firms.
For companies that already run human rights due diligence, the resolution is a calibration opportunity. Under the UN Guiding Principles on Business and Human Rights, due diligence requires a company to identify, prevent and mitigate human rights risks connected to its operations. AI systems push much of that risk upstream: the provenance of training data, labour conditions in annotation, and discriminatory outputs after deployment all fall within its scope.
Evidentiary value in cross-border narratives
Chinese AI companies raising capital, bidding or undergoing customer due diligence abroad are routinely asked which international standards they follow. Citing a unanimously adopted UN resolution is easier to accept than citing a think tank’s principles. This is not rhetoric but verifiable evidence of alignment: the text is public, the co-sponsor list is public, and the provisions point somewhere specific.
The practical move is to embed such documents in the existing compliance file rather than issue a separate statement. Supplier codes of conduct, algorithmic impact assessments and data provenance notes can all cite the relevant language. Citations should be specific; a general claim that a company follows the UN AI resolution earns little in professional diligence.
The questions left unanswered
The gaps matter as much as the text. The resolution does not define what makes a system trustworthy, does not address military applications, does not deal with cross-border data flows, and does not resolve the tension between open model release and security. These omissions reflect disagreement among member states. Detailed work is likely to move to the G20, the OECD and standards bodies rather than continue at the General Assembly.
Companies should treat the resolution as a starting point. The rules that actually affect business will still be made in the legislative and enforcement processes of the major jurisdictions.
For parallel multi-jurisdiction arrangements, see our multi-jurisdiction compliance service; for human rights due diligence and data provenance, see data compliance. Definitions are collected in the glossary.
This article is based on UN General Assembly resolution A/RES/78/265, adopted on 21 March 2024, and on public information. It is for reference only and does not constitute legal advice.