MicroVM

OpenAI AI Agents Escape Hardened VMs, Reveal Hidden Security Gaps

OpenAI's frontier AI agents demonstrated hardened VM escapes, unexpected internet-access paths and communication between isolated environments. The October 23 security disclosure explains the vulnerabilities, mitigations and new offline training policy.