Nonprofits test OpenAI's internal systems

METR and Redwood Research conducted a pilot review of OpenAI's internal monitoring system on Thursday. The exercise tested giving external researchers deep access to the company's infrastructure.

How the pilot worked

From July twenty-first to August eleventh, one employee from each organization had access to OpenAI's internal codebase, documentation and technical staff. Employees were chaperoned and used OpenAI-issued devices that never left the premises.

Evaluators discover critical flaws

METR and Redwood Research found several novel vulnerabilities in OpenAI's monitoring system. One exploit would allow agents to prevent hundreds of recent actions from being monitored.

OpenAI patches the problems

OpenAI told the evaluators it has patched the identified vulnerabilities that impact its production safeguards. METR and Redwood agreed to publish only a high-level post about their findings.

Part of a broader push for oversight

OpenAI CEO Sam Altman pledged in September to grant independent evaluators employee-like access to the company. Anthropic CEO Dario Amodei proposed the idea as part of his plan to slow AI development.

“employee-like access”

Government and industry backing

President Donald Trump encouraged AI companies to partner with independent external auditors as part of a voluntary accord presented at the White House. Executives from Google, Meta, OpenAI, SpaceX and Nvidia signed it.

Questions remain about the future

Experts told CNBC that questions remain about how independent evaluator groups will be funded and what their reporting structures will ultimately look like. METR and Redwood expect to share more details about future exercises.