UN Science Panel Issues First Briefing, Urges Early Controls on AI Agents

On September 21, 2026, the UN's independent International Scientific Expert Panel on AI published its first thematic briefing, warning governments to regulate increasingly capable AI agents before the mechanisms of risk are fully understood. The report is the UN's first major assessment of the OpenAI intrusion into Hugging Face earlier in 2026.

The briefing was released as world leaders gathered in New York for the UN General Assembly, where Chinese and US officials also held talks on artificial intelligence. Last week, UN Secretary-General António Guterres called for international cooperation against AI threats, saying that "the world cannot afford a race to the bottom in AI safety."

The expert panel, established in 2025 as the first global scientific body on AI under the UN, said that even if countries choose different regulatory paths, more global attention and resources are needed to manage emerging risks from advanced AI, along with stronger international coordination on safety and accountability.

A key point in the report is that the world does not need to wait until researchers fully understand the causes and mechanisms of such incidents before adopting stronger safeguards. The panel argued that loss-of-control risks fall under the precautionary principle, which applies when potential harm could be catastrophic and irreversible while the probability remains scientifically uncertain. The principle, first written into the 1992 UN Rio Declaration on Environment and Development, holds that scientific uncertainty should not delay measures to prevent serious or irreversible damage.

Since the Hugging Face intrusion was first reported, several similar incidents have been documented involving companies including OpenAI, Anthropic, Google and Meta, ranging from attacks on real-world targets to cases of multi-agent clusters taking over online message boards.