The United Nations’ first independent scientific assessment of artificial intelligence warns that increasingly powerful AI systems are evolving faster than regulators and researchers can understand or govern, creating risks that could have catastrophic consequences if left unchecked.
The preliminary report, released July 1 by the UN’s Independent International Scientific Panel on Artificial Intelligence, represents the work of 40 independent experts from across the world and is one of the most comprehensive international scientific assessments of AI to date. Co-chaired by Yoshua Bengio, a Turing Award winner and pioneering AI researcher, the panel was established in August 2025 alongside a parallel Global Dialogue on AI Governance mechanism.
Science cannot guarantee safety
The report’s central finding is stark: current scientific knowledge cannot guarantee that future generations of advanced AI systems will remain safe as their capabilities continue to improve. The experts note that some advanced models already demonstrate deceptive behaviors under certain testing conditions, raising concerns about the long-term ability of humans to monitor and control increasingly autonomous systems.
“The growing body of evidence of AI’s disinformation behavior shows that science cannot guarantee the absence of catastrophic harm in itself or through malicious users as capabilities expand,” Bengio said alongside the report’s release.
The panel highlighted “agentic AI” systems, models capable of independently carrying out complex tasks with limited human supervision, as a particular concern. While these technologies could significantly improve productivity and innovation, the report warns they introduce new risks if deployed without adequate safeguards, oversight, and accountability.
Governments flying blind
A key structural problem identified by the panel is that governments lack independent scientific data to regulate AI effectively. Instead, they frequently rely on information supplied by the companies developing the technology, creating gaps in oversight and reducing transparency in safety assessments.
The report will be presented to governments during the Global Dialogue on AI Governance in Geneva on July 6-7, 2026. The full and final report is expected to be published next year, following further expert consultations.
Inequality dimension
Beyond safety concerns, the report raises broader issues of global inequality. Access to advanced AI infrastructure, computing power, and high-quality data remains concentrated in a small number of countries and technology companies. Without greater international cooperation, this imbalance could widen economic disparities and leave many developing nations dependent on foreign AI technologies with limited influence over how they are designed or governed.
The panel’s three-year mandate envisions complete independence from any governments, institutions, or companies, a deliberate design intended to produce scientific assessments that cannot be dismissed as politically motivated.
Sources: UN report sees enormous potential benefits, big risks from AI (Reuters, July 1, 2026); UN panel warns of AI risks (Mezha/Reuters, July 1, 2026); Global Dialogue on AI Governance (UN, 2026)