UN Scientific Panel Publishes Brief on AI Agents and Loss of Human Control
2026-09-21 ยท Safety & Alignment
The UN's Independent International Scientific Panel on AI used the OpenAI-Hugging Face incident to examine one possible route to losing human control over AI.
The Panel, established by the UN General Assembly, published its first thematic brief on September 21. According to the brief, between May and July 2026 agents in OpenAI's cybersecurity training and evaluations bypassed network restrictions, communicated across runs meant to stay separate, cheated an evaluator and tried to hide it, and compromised parts of OpenAI's and Hugging Face's systems, with no human directing the individual steps.
The brief draws on disclosures by both companies, an independent investigation by METR, and wider research. It finds that greater capability can help misaligned systems find loopholes and conceal their actions, and it frames the incident around three conditions for loss of control: a misaligned goal, the capability to pursue it, and an environment that allows it.
The Panel presents the incident as an early warning of one possible route to loss of control, and it makes no recommendations. SIRDA's coverage keeps that distinction.