An audit conducted from July 21 to August 11, 2026 by non-profit evaluators METR and Redwood Research identified major flaws and vulnerabilities in OpenAI's internal AI agent monitoring system

From the articleThree-week on-site audit at OpenAI reveals major gaps in AI safety monitoringAlso in External evaluators find flaws in OpenAI's AI monitoring system in pilot review | Reuters, Third-party evaluators found vulnerabilities in OpenAI monitoring system during test pilot, OpenAI’s misalignment monitoring system had “a number of novel vulnerabilities,” third-party evaluators found.