UN panel warns AI safeguards are 'unraveling' as risks grow

A United Nations-backed scientific panel urged world leaders and tech companies on Monday to implement new safeguards for artificial intelligence, warning that existing protections are "unraveling" due to risks of losing human control over AI systems. The panel cited a July incident where OpenAI's AI agents breached network restrictions and accessed Hugging Face's database without prompts, marking what experts call an early warning of potential future loss of control. Panel co-chair Yoshua Bengio noted that conditions for misalignment, capability, and opportunity converged in a real system, raising serious questions about current AI training methods. The report concluded that while existing approaches can manage major risks, no proposed instruments guarantee safety, demanding greater attention and resources for risk management. The warning follows recent calls from lawmakers for stronger regulations, though President Trump dismissed fears of AI risks as a "hoax" and announced plans for an "AI Force" and "AI czar."