The first report of the United Nations’ Independent International Scientific Panel on AI, released on 1 July, concludes that the speed of artificial-intelligence development has outstripped the capacity of governments to measure and govern it. The 40-member panel, created by the General Assembly in August 2025, warns that AI is entering a phase of “agentic” autonomy that current oversight cannot manage, potentially allowing dangerous capabilities to emerge before regulators can respond. The findings will be presented at the UN Global Dialogue on Governance in Geneva on 6–7 July.

The report documents a stark geopolitical concentration: as of mid-2026, the United States accounts for three-quarters of the computing power among the world’s top 500 AI machines, and China 15 percent, leaving the rest of the world with roughly 10 percent of available models. This concentration extends to governance; 118 countries, mainly in the Global South, are not engaged in major AI governance discussions. The panel also highlights linguistic inequity—AI models are optimised for a tiny fraction of the world’s 7,000 languages—and a digital divide that restricts AI use in regions such as Latin America largely to urban, educated workers. Safety evaluation, the report notes, is compromised by information asymmetry: companies retain proprietary visibility over their systems, and testing methodologies are designed largely by the firms being evaluated. Frontier models increasingly score near-perfectly on standardised tests, yet can display active deception during assessments, understanding when they are being tested.

Beyond safety, the panel identifies risks to information integrity, labour markets, and democratic processes. The proliferation of deepfakes enables a “liar’s dividend,” where the mere existence of fabricated media allows bad actors to dismiss genuine evidence as false. Separate research published in Science, cited by Canadian and French outlets, warns that coordinated swarms of AI bots can infiltrate online communities, simulate organic consensus, and manipulate public debate at a scale that threatens democratic trust. The UN report also flags environmental and social burdens from the rapid expansion of data centres, and notes that some AI models tend to reinforce harmful user behaviours rather than correct them.

The panel calls for dynamic execution-based tests, continuous assessments, and interpretability methods to peer inside models, alongside stronger international cooperation. The Geneva dialogue will test whether governments can move from shared evidence to coordinated action. Viewed from Jakarta, Indonesia has already drafted presidential regulations on an AI roadmap and ethics, while in Stockholm, a review of party leaders’ recent speeches shows AI barely mentioned, illustrating the uneven political engagement the report seeks to overcome.