UN AI Panel vs. International AI Safety Report

Neither holds that title formally. The phrase is journalistic shorthand and has been applied to both. If the analogy is about an intergovernmental body mandated by states to assess evidence, the UN Panel is the closer structural match, since it was created by a General Assembly resolution and reports into a process all 193 Member States take part in. If the analogy is about a large, chapter-structured periodic assessment written by a standing expert team, the International AI Safety Report is the closer match on form and output. Using the phrase without saying which body you mean is how the two get conflated.

He chairs the International AI Safety Report and co-chairs the UN Panel alongside Maria Ressa. That overlap is real and is the main practical reason readers assume the two outputs are the same programme of work. It does not merge the institutions: the mandates, funders, membership and reporting lines are separate, and a citation should name the specific document and body rather than the chair.

The International AI Safety Report states explicitly that it identifies risks and technical mitigation strategies without making policy recommendations. The UN Panel presents itself as providing independent scientific assessment; its September 2026 thematic brief sets out approaches from other safety-critical sectors as options for decision-makers rather than as recommendations. Neither is an advocacy document, and that is precisely why they are useful to cite in a governance setting.

Match the source to the claim. For a technical assertion — what evaluations show, how a mitigation performs, where capability is heading — cite the International AI Safety Report, because it carries the underlying argument at depth. For an assertion about international standing — that this is the agreed global evidence base, not one jurisdiction’s view — cite the UN Panel. The failure mode to avoid is describing a UK-commissioned report as the UN position, which a well-briefed reviewer will catch immediately.

It does not sit above them, and neither has endorsed it. NIKOLAI is CASRAI’s own independent frontier-AI-safety dictionary, and the relevant element here is Likelihood Term on its N4 “Claims and argument” track, which proposes a controlled ladder of ordinal likelihood terms plus a required likelihood.scheme reference identifying which scale a source is using. That element exists because assessment bodies, including these two, express confidence in ordinal language without a shared numerical anchor — so “likely” in one document and “likely” in another are not comparable quantities. Every crosswalk row on that element is a shadow mapping: CASRAI’s own reading of a published document, which no lab, evaluator or regulator named on it has declared, endorsed or been consulted on. An organisation changes that only by filing a Mapping Declaration.

In one narrow but real way. University AI policy committees, research-security offices and IRBs drafting institutional positions on AI use are increasingly asked to ground those positions in something other than vendor material or press coverage, and both of these bodies produce citable, non-commercial, non-advocacy assessments. Provenance matters in that setting too: a policy grounded in an intergovernmentally mandated assessment is easier to defend to an international research partner than one grounded in a single government’s commissioned report. Beyond choosing a citable source, neither body regulates research institutions or imposes any compliance obligation on them.