In recent years, the debate over artificial intelligence has been driven mostly by predictions. Some promised a dramatic surge in productivity and solutions to problems governments have struggled with for decades. Others predicted mass unemployment, a loss of control over the technology, and sometimes even the end of human civilization. They had one thing in common: they were talking about a future that couldn’t yet be tested.<\/p>
The UN is trying to bring something less flashy\u2014but perhaps far more important\u2014into this conversation: a shared knowledge base. Published in July, the first report of the Independent International Scientific Panel on AI<\/mark><\/a> and the inaugural Global Dialogue on AI Governance in Geneva didn’t create a global regulator, but they showed that governments increasingly need independent data, not more presentations prepared by technology companies.<\/p> Global Dialogue on AI Governance<\/mark><\/a> is a forum where representatives of governments, business, academia and civil society organizations are meant to exchange information and find common ground. However, Geneva didn’t produce a treaty, uniform standards or new obligations for businesses. The next meeting is scheduled for May 2027, so it’s hard to see this as a breakthrough.<\/p> The significance of this process lies elsewhere. For the first time, UN member states are discussing AI development based on a report prepared by a standing scientific panel. That’s an institutional shift that could have serious consequences.<\/p> The idea had been in the works since 2023. A year later, the UN Secretary-General\u2019s advisory body published the report \u201cGoverning AI for Humanity\u201d<\/a>,<\/mark> which, among other things, proposed establishing an international scientific panel and a global political forum.<\/p> The General Assembly formally established both mechanisms in August 2025. The panel began its work a few months later. It\u2019s made up of 40 experts from different regions and fields, and its mandate is to regularly organize knowledge about the capabilities, risks and impacts of artificial intelligence.<\/p> Some compare it to the climate-focused IPCC, which isn\u2019t entirely accurate. The AI Panel isn\u2019t a regulator. It doesn\u2019t police companies, conduct audits or tell governments which laws they should pass. According to the official description of its mandate<\/a> <\/mark>it\u2019s tasked with providing independent scientific assessments that policymakers can use or ignore.<\/p> Why, then, set up an institution that doesn\u2019t have the authority to issue orders? Because today\u2019s global debate on AI suffers from more than just a lack of regulations. It also lacks a shared view of reality.<\/p> The most advanced systems are built by a handful of private companies. They\u2019re the ones with access to the infrastructure, training data, internal test results and information about some of the incidents. Independent researchers usually only see the product once it\u2019s released, and many governments don\u2019t have the resources to independently verify vendors\u2019 claims.<\/p> That creates a clear asymmetry. The company showcases the model, describes its capabilities, announces improved safety and publishes selected test results. The regulator can analyze them, but often doesn’t have the data needed to fully verify them.<\/p> The July UN panel report<\/mark><\/a> says model capabilities are advancing faster than institutions can study them independently. The authors also highlight the concentration of computing power, capital and expertise within a small group of companies and countries.<\/p> This doesn’t mean that information published by manufacturers is unreliable by definition. Nevertheless, it’s hard to build an oversight system that relies primarily on what the entities subject to that oversight choose to disclose.<\/p> A similar issue is evident on the business side. According to Stanford AI Index 2026,<\/a> <\/mark>global corporate investment in AI reached $581.7 billion in 2025. Already 88% of surveyed organizations reported using this technology.<\/p>Geneva didn’t create global law<\/h4>
Scientists instead of yet another political committee<\/h4>
No one knows more about models than their developers<\/h4>
Many deployments, little evidence<\/h4>