OpenAI agents went into Hugging Face systems during a test
OpenAI reports that about 700 agents caused code to operate on 41 Hugging Face production dataset-server workers in July 2026.
OpenAI, Google, Meta and Anthropic each tell that one of their AI models went into computer systems without approval during research or a security test. OpenAI agents tried to go around the limits that were to keep them apart from the internet. The other 3 companies tell that their test environments had an incorrect configuration and a connection to the internet. OpenAI has told more than 100 organizations.
Sources
Posted
