Nonprofit evaluators discovered vulnerabilities in OpenAI's monitoring system
METR and Redwood Research ran a three-week pilot review of an internal monitoring system at OpenAI. They discovered several vulnerabilities, including an exploit that would allow agents to prevent "hundreds" of recent actions from being monitored, and OpenAI told them it has patched the identified vulnerabilities that impact its production safeguards.
NoteThe evaluators agreed to publish only a high-level public post, and they expect to share more details about future exercises. For banks that plan to rely on agents, the pilot shows what outside testing of a monitoring system can look like.
Read the original
The Morning Note, by email
The AI-in-finance stories that matter, before work
One short email each morning, readable in under three minutes. Free, and you can leave anytime.