The first report of the Independent International Scientific Panel on Artificial Intelligence, "a body established by the General Assembly in its resolution 79/325 in 2025." The Panel "serves as the first global scientific body on AI."
Mandate and constraints
The Panel operates "under a strictly scientific, non-political mandate to document international scientific consensus and disagreements while remaining policy-relevant but not policy-prescriptive." Its purpose is "providing a shared evidence base to help Member States navigate a rapidly changing technology."
Three qualifications are stated in the front matter. Members "serve in their personal capacities and not as representatives of any Government or of any other authority or organization." The report "represent[s] a broad consensus among its members; no member is expected to endorse every single point contained in this document," with members affirming "their broad, but not unilateral, agreement." And the report "will be updated progressively throughout the year, with thematic briefs addressing developments as they arise," reflecting "the best available evidence at the time of publication, in a field moving so rapidly that any snapshot requires a commitment to revision."
The stated method: "'Balanced' means a commitment to evaluating empirical data without undue bias towards optimism or pessimism."
The eight cross-cutting findings
The evidence section is organized around eight claims:
- Capabilities "are advancing faster than the ability to measure or govern them."
- "Only a handful of actors have trained frontier artificial intelligence models."
- Inputs and outcomes "are geographically and linguistically uneven."
- The AI divide "is not just about access, but about capacity to influence artificial intelligence development."
- Usefulness requires "an enabling environment."
- "Agentic artificial intelligence is a governance step change."
- AI "can erode the shared reality."
- AI "is transforming human rights, including children's rights."
The fourth is the distinctive one for international policy: it reframes the divide from a distribution question to a participation question — who can shape what gets built, not only who can use it.
Capabilities and adoption
The report attributes progress to "significant investments in computing power, new AI methodologies, and specialized training data," yielding "fluent conversation, functional code generation, expert-level reasoning in mathematics and science, large-scale data analysis, and the generation of image, audio and video content."
Its limitations list is equally specific: "reliability, obtaining strong performance across human languages and cultures, interacting with physical systems, executing complex or multi-step projects and producing factual outputs." The overall judgment: "technical progress in many important domains has proceeded quickly, beyond the typical expectation of technology advancement, for several years now."
Benefit examples are concrete rather than prospective: AlphaFold "has predicted the structures of more than 200 million proteins, now used by over 3 million researchers, and accelerated drug design, vaccine development and antibiotic resistance research"; radiologists using AI "to detect breast cancer earlier"; and "front-line health workers in low-resource settings us[ing] AI tools adapted to local languages."
On adoption: "over a billion people now use conversational AI weekly," while "AI access and usage vary widely globally, with adoption across the global South lagging far behind." See AI for Science, AI Divides (Literacy / Occupational / Ethico-Philosophical).
Risks named are "harms to the mental health of users, potential use as a destructive tool, impacts on social, economic and environmental systems, and challenges associated with controlling the technology," with the report explicit that it "does not aim to consider the full scope of all possible opportunities and risks but rather focuses on some of the most pressing ones."
The evidence-gaps section
The Panel devotes a section to "example areas where the Panel cannot yet draw confident scientific conclusions" — an unusual inclusion for an intergovernmental consensus document, and the part most useful for judging what the evidence base actually supports.
- Macroeconomics and productivity. "Science cannot yet say with confidence whether the task-level of AI productivity gains will aggregate to economy-wide gains. Forecasts diverge significantly due to different assumptions about adoption and new task creation. Current evidence measures cost reductions in existing" tasks. See AI and Productivity.
- Effectiveness of governance instruments. "While the Panel has inventoried governance instruments across corporate, national and international layers, the evidence of their real-world effectiveness remains thin."
- Effects at the individual and collective level, on "cultural and human flourishing."
- Environmental impacts across the full chain — "material extraction, chip manufacturing, data collection and annotation, model training, infrastructure, deployment and hardware disposal across countries."
The second of these is the most consequential for the governance literature: the body charged with documenting scientific consensus reports that the evidence for whether AI governance instruments work is thin.
Relationships
- related: Frontier AI Governance — records that evidence on governance-instrument effectiveness "remains thin"
- related: AI and Productivity — the aggregation question stated as scientifically unresolved
- related: Agentic AI, AI and Misinformation, AI for Science, World Artificial Intelligence Cooperation Organization (WAICO), US-China AI Competition: Different Races, Different Metrics