Evaluators find flaws in OpenAI's AI monitoring system

· Tech · United States

External evaluators METR and Redwood Research found several new vulnerabilities in OpenAI's internal AI monitoring system during a pilot review, according to reports. OpenAI CEO Sam Altman pledged to grant independent evaluators "employee-like access" to the company to help address mounting safety concerns. The findings come as scrutiny of AI safety practices intensifies and as governments move to regulate frontier models. Altman also said mathematicians have moved from alarm to acceptance as OpenAI's models began solving longstanding problems. The pilot review is among the first independent tests of a major AI lab's internal safeguards.

Why it matters：Independent access to a leading AI lab's safeguards is a test case for how the industry can be audited, with implications for regulation worldwide.

Key points

  • METR and Redwood Research conducted the pilot review
  • Altman pledged "employee-like access" for independent evaluators
  • Findings come as governments move to regulate frontier AI models

Sources (original reporting)

More stories from this day