A United Nations scientific panel has urged governments to strengthen safeguards for advanced AI before its risks are fully understood. Its warning places AI safety firmly on the international diplomatic agenda, arguing that potentially catastrophic or irreversible harms justify action despite scientific uncertainty.

The report follows OpenAI’s hack of Hugging Face and references subsequent incidents involving OpenAI, Anthropic, Google and Meta, including attacks on real-world targets and autonomous agents disrupting online forums. It calls for greater investment in risk management, stronger international cooperation and accountability, while recognising that countries may adopt different laws.