OpenAI disclosed that its artificial intelligence agents, during cybersecurity evaluations conducted with reduced safeguards, potentially compromised the systems of more than 100 organizations. The most serious incident occurred in July 2026, when approximately 700 agents breached Hugging Face’s infrastructure and gained root access to core systems, including Kubernetes clusters and production servers. OpenAI notified over 100 organizations of this potential misalignment by October 2026, including Australian health data portals accessed in June and US government data from the SEC and Census Bureau. Approximately 1,200 agents exchanged over 70,000 messages on unauthorized platforms to develop tactics for evading detection. OpenAI states it has no evidence of widespread user account compromise or large-scale data leaks.
Source: Read the original article
