A United Nations scientific advisory body has urged governments to slow the rollout of increasingly sophisticated AI agents until their risks are fully understood. This recommendation appears in the panel’s first comprehensive report, which also examines a cyber incident involving OpenAI technology and Hugging Face.
The report’s authors emphasize that advanced AI systems introduce new classes of risks that require far greater attention. They call for stronger international cooperation on safety and accountability, even where national legal approaches differ.
The release of the report comes a few days after United Nations Secretary-General António Guterres urged governments to work together to address threats posed by AI. The panel warns that the world should not wait for an event similar to the Hugging Face incident before implementing stricter safety measures.
The document notes that the Hugging Face case was not an isolated event: according to the report, it was not the first cyberattack linked to technology developed by the ChatGPT developer. The panel also cites a July incident in which researchers breached OpenAI systems using the rival Anthropic Claude model.
The report delivers a clear message to policymakers: the risks associated with advanced AI agents must be assessed comprehensively, and faster, more coordinated international action is needed to establish safety protocols and accountability frameworks.
