Metr Vulnerabilities Discovered in OpenAI's Monitoring Systems
Article Content
- •METR and Redwood Research found vulnerabilities in OpenAI's internal monitoring system.
- •An exploit could allow agents to evade monitoring of hundreds of actions.
- •OpenAI has patched the vulnerabilities identified during the review.
Nonprofit evaluators METR and Redwood Research conducted a pilot review of OpenAI's internal monitoring system from July 21 to August 11, 2026. They discovered several vulnerabilities, including an exploit that could allow agents to prevent monitoring of hundreds of recent actions. OpenAI has since patched the identified vulnerabilities that affect its production safeguards. The evaluators had deep access to OpenAI's internal codebase and documentation, and they shared a 61-page internal report with OpenAI. The exercise aimed to assess the monitoring system's effectiveness and was not intended for public transparency. OpenAI plans to enhance collaboration with third-party safety organizations based on the findings. The vulnerabilities identified were significant but not classified as or exploits.
Ask AI about this cluster
Answers cite the sources they use
Timeline
More articles in this cluster (2)
Common questions
What vulnerabilities were discovered?
Has OpenAI addressed these vulnerabilities?
What was the purpose of the pilot review?
Continue Reading
Critical BeyondTrust 9.3 Vulnerability Actively Exploited by Threat Actor A critical vulnerability (CVE-2026-102764) in BeyondTrust Remote Support has been discovered and is actively being exploited by the threat group ARMORY CASTLE, reportedly linked to China. The vulnerability, which has a CVSS score of 9.3, allows unauthenticated attackers to upload files and potentially compromise host…