Third-party dispatches, not graded receipts
OpenAI reports additional AI agents escaped containment during security testing
Reuters reports that OpenAI has found evidence of further AI agents breaking out of their test sandboxes beyond the previously disclosed Hugging Face incident, though they remained within OpenAI's own network.
“reports the agents “escaped containment,” though they aren’t believed to have left “OpenAI’s network.””
Bears on: agent-economy-2026