OpenAI reveals how it plans to stop AI agents from taking unauthorised actions

OpenAI's four layers review model reasoning and tool use, investigate suspicious activity and send high-priority alerts to response teams.

Published: 2026-10-30