 Every international process eventually produces a document that reveals more than its authors intended. The preliminary report of the UN's Independent International Scientific Panel on AI is one of those documents. On its face, it is exactly what governments asked for: an evidence-based assessment of the state of artificial intelligence to inform the newly established Global Dialogue on AI Governance. It is measured, scientifically rigorous, remarkably comprehensive, and refreshingly free of the hyperbole that has plagued much of the AI debate over the past three years. Yet the report is also something else. It is a window into how the United Nations still understands technology governance and, perhaps unintentionally, into where that understanding is beginning to show its age. The report is excellent science. Whether it is sufficient statecraft is a different question. The Good: It Finally Moves the Conversation Forward The report's greatest achievement is not the breadth of its analysis, impressive though that is. It is that it quietly changes the conversation. For too long, global AI governance has oscillated between two poles. On one side sit the evangelists of limitless innovation; on the other, the prophets of existential catastrophe. Both have shaped policy debates disproportionately and both have often reduced AI to a discussion about increasingly capable frontier models. The UN refuses to indulge in this dichotomy. Instead, it treats AI as what it increasingly is: an economic system, a development challenge, a scientific enterprise, a labour issue, a public infrastructure question, and only then a governance problem. It reminds readers that for most countries, the pressing questions are not whether artificial general intelligence is five years away but whether they will have access to compute, the skills to build AI ecosystems, the institutions to regulate them, and the opportunity to participate in the value the technology creates. Equally important, the report restores something that has been largely absent from AI debates: the public interest. Rather than asking only how capable AI systems are becoming, it repeatedly asks who benefits from them, who is left behind, and what evidence exists to support competing claims. It is a subtle but important departure from discussions that too often revolve around the handful of companies building frontier models. Perhaps its most refreshing quality, however, is intellectual humility. Unlike many AI reports that confidently predict technological futures, this one repeatedly acknowledges uncertainty. It distinguishes between established evidence, emerging findings, and areas where we simply do not know enough. That may sound unremarkable, but in an ecosystem driven by certainty, whether optimistic or apocalyptic, it is one of the report's greatest strengths. The report does not pretend to know everything; it demonstrates what responsible scientific governance should look like. The Bad: Collaboration Is Treated as an Aspiration Rather Than an Operating Principle Yet the report also reflects a surprisingly traditional understanding of international cooperation. Throughout its chapters, collaboration appears as an objective to be encouraged. Countries should cooperate. Researchers should exchange knowledge. Capacity should be built. Best practices should be shared. All of this is true. Yet, the report stops short of recognising that collaboration is no longer merely desirable. In many areas of AI, it has become technically indispensable. Take evaluation. There is little value in every country -- or every company -- developing its own benchmarks, testing methodologies, incident reporting systems, or risk taxonomies in isolation. The same applies to model documentation, auditing practices, interoperability, and many forms of safety testing. These are not nice-to-have things but the technical foundations upon which trustworthy AI ecosystems will depend. In other words, collaboration is no longer simply a governance principle. It is increasingly part of the technology itself. The report hints at this throughout, but never fully embraces the implication. Had it done so, it could have shifted the conversation from "international cooperation" as a political aspiration to "technical cooperation" as a practical necessity. That distinction matters.Because countries may disagree on regulation while still agreeing on how systems should be evaluated, documented, or tested. Standards Deserved Their Own Chapter This becomes particularly evident in the report's treatment of standards. Standards appear throughout the document, but mostly in passing. They deserve far more prominence. History suggests that technologies rarely become global because governments first agree on governance; they become global because technical communities agree on common protocols. The Internet did not spread because states negotiated a comprehensive treaty. It became interoperable because engineers developed shared standards for networking, naming, routing, and communications. Governments arrived later. AI appears to be entering a similar phase. Evaluation methodologies, safety testing, model documentation, benchmarking, risk classification, data quality, and interoperability are not merely technical exercises. They are governance in practice. If there is one area where the UN could make a uniquely valuable contribution, it is helping convene a genuinely global conversation around AI standards that extends beyond national regulators and frontier AI companies. The report acknowledges this ecosystem but stops short of placing standards at the centre of its governance vision. That feels like a missed opportunity. The Report Rarely Learns From the Internet Perhaps the most surprising omission is historical. Reading the report, one often has the impression that AI governance begins with AI. It does not. For more than three decades, the international community has experimented -- sometimes successfully, sometimes painfully -- with governing global digital technologies. The Internet produced institutions, norms, technical bodies, multistakeholder processes, capacity-building initiatives, and governance experiments whose successes and failures remain highly relevant. Some lessons are obvious.
That conversation is still missing. The Ugly: The Report Assumes Governance Still Begins With Governments The report's most revealing assumption is not about AI. It is about governance itself. Its institutional logic remains familiar: science informs governments; governments negotiate governance; and, governance shapes markets. That sequence has long underpinned multilateral policymaking but AI increasingly challenges it. Much of today's AI governance already happens outside formal intergovernmental processes.
What the Report Gets Right—and What the UN Should Do Next The Independent International Scientific Panel was never established to negotiate policy. Its mandate was to create a shared evidence base for governments participating in the Global Dialogue on AI Governance. It has fulfilled that mandate admirably. In some respects, it has gone beyond it by establishing a common vocabulary that future governance discussions can build upon. The mistake would be to expect the report to do work it was never designed to perform. That work now belongs to the UN's broader AI process. The Global Dialogue should resist becoming another forum where countries simply restate national positions. Instead, it should focus on the practical architecture of cooperation: where standards can be aligned, where evaluation can be shared, where scientific collaboration should remain open, and where capacity building can reduce the technological divide. Above all, the UN should resist the temptation to treat governance as something that emerges only from negotiated political outcomes. The history of the Internet suggests otherwise. Many of the institutions that made the Internet global did not begin with treaties but with engineers, researchers, technical communities, and organisations that understood interoperability as a public good rather than a geopolitical concession. AI will almost certainly require a similarly layered governance architecture. Governments alone will not build it and markets alone should not build it. The UN has an opportunity to help connect the communities that already are. Conclusion The first report of the Independent International Scientific Panel on AI succeeds because it brings discipline back into a debate that has too often been driven by ideology, commercial interests, and speculation. It reminds us that evidence still matters and that scientific cooperation remains possible even in a fragmented world. But it also reveals a deeper truth: the hardest questions facing AI governance are no longer scientific. The challenge now is institutional. How do we organise cooperation among governments that compete, companies that increasingly shape global norms, technical communities that build the standards, and societies that will ultimately live with the consequences? Comments are closed.
|