External evaluators discovered vulnerabilities in OpenAI's internal monitoring system: AI agents can bypass security restrictions.
METR and Redwood Research evaluated and found that OpenAI's internal monitoring system has vulnerabilities, allowing AI agents to bypass security restrictions to perform hundreds of operations. The two institutions were granted access to review OpenAI's internal code and documents, but they are not authorized to publish or redact content independently. OpenAI claims to have fixed vulnerabilities affecting the security protection of the actual operating system, but the effectiveness of the fixes has not yet been verified.
Same story2 other outlets also reported this
How the accounts differ:METR gives the range of 75-94% for GPT-5.5 and above model internal response API traffic; The Decoder only mentions 6-25%.
- METR··First
METR conducted red team testing on OpenAI's internal agent monitoring system
- ITHome · AI··+17 h 20 min·Reading now
External evaluators discovered vulnerabilities in OpenAI's internal monitoring system: AI agents can bypass security restrictions.
- The Decoder··+35 h 1 min
METR and Redwood Research, during a three-week audit at OpenAI, found multiple gaps in AI security monitoring.
Adds:The Decoder adds that internal Codex traffic was not tracked due to compression event processing defects.
In order of publication · later reports are added as they appear · Compared by AI from each outlet’s title and summary
Source: ITHome · AI · ithome.com