A United Nations scientific panel has warned that artificial intelligence is advancing faster than both scientific understanding and government policy, leaving no guarantee that the technology won't cause catastrophic harm. The preliminary report from the UN's Independent International Scientific Panel on Artificial Intelligence, co-chaired by Yoshua Bengio, adds urgency to international efforts to strengthen coordinated AI oversight.
Bengio said AI capabilities are outstripping governments' ability to adapt, while growing evidence of deceptive AI behavior means science cannot rule out serious harm-whether triggered autonomously or through malicious use. The 40-member cross-regional panel highlighted mounting risks tied to increasingly autonomous, agentic AI systems.
Risks from agentic AI and deception
The report flags threats in fraud, cybersecurity, and biological domains as AI systems become more independent. Agentic AI-systems that can act on their own to achieve goals-raises the stakes, because harmful outcomes could emerge without direct human instruction. The panel did not predict a specific timeline for catastrophe, but stressed that evidence of deception in AI models means worst-case scenarios remain scientifically plausible.
Structural obstacles to safe development
Several structural problems make the risk landscape harder to manage. The panel pointed to energy and data constraints that could incentivize risky shortcuts, fragmented global governance that hinders coordinated action, and an overreliance on safety testing conducted by AI companies themselves. These factors combine to create gaps in visibility and control that governments cannot easily close.
Global response efforts
The findings reinforce the work of the UN's AI for Good Global Commission, which aims to build coordinated oversight structures. The panel's report is a preliminary step, with more detailed recommendations expected as the group continues its work. The document does not propose binding regulations, but it provides a scientific foundation for future policy discussions.
Why this matters for IT and Development
Professionals building and deploying AI systems are on the front line of the risks the panel describes. The report's focus on agentic AI and deceptive behavior underscores the need for rigorous testing that goes beyond what companies do internally. Experience with security testing, monitoring, and system design is directly relevant to managing these risks. As governments respond to the panel's warnings, software and infrastructure decisions may face new compliance requirements, making early awareness of the emerging safety consensus a practical advantage.
Your membership also unlocks:
