OpenAI's internal monitoring system found vulnerabilities in third‑party pilot
Nonprofit evaluators METR and Redwood Research ran a three‑week pilot from July 21 to Aug. 11th to test OpenAI's internal misalignment monitoring system. Employees from each group were chaperoned on site, used OpenAI‑issued devices that never left the premises, and accessed code, documentation and some technical staff.
Key points
- METR and Redwood Research conducted a three‑week pilot from July 21 to Aug. 11th reviewing OpenAI’s internal misalignment monitoring system.
- Evaluators found several novel vulnerabilities, including an exploit that could stop hundreds of recent actions from being monitored.
- OpenAI said it patched the vulnerabilities and shared a 61‑page internal report with the evaluators.
The teams reported several novel vulnerabilities, including an exploit that could prevent hundreds of recent actions from being monitored. OpenAI told the evaluators it has patched the identified issues and shared a 61‑page internal report with them. OpenAI CEO Sam Altman pledged in September to grant independent evaluators "employee‑like access," a suggestion originally made by Anthropic CEO Dario Amodei. President Donald Trump also urged AI firms to partner with external auditors in a voluntary White House accord signed by leaders from Google, Meta, OpenAI, SpaceX and Nvidia. METR and Redwood said they will publish more details about future exercises as the collaboration continues.
The story so far
6 episodes →- OpenAI's internal monitoring system found vulnerabilities in third‑party pilotthis story
Third-party evaluators find vulnerabilities in OpenAI monitoring system during test pilot
CNBC Technology · 22 October 2026
Loading the full article…
This text was published by CNBC Technology and written by Ashley Capoot. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
Coverage and discussion
2sources- OpenAI let METR and Redwood Research inside to test its agent monitoringPress · cryptobriefing.com ·
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Policy & Regulation
All →- Nvidia CEO Jensen Huang pledges $2 billion for Europe’s sovereign AI push · 8 src
- OpenAI quietly withdraws from pushing for chip export restrictions to China, report says · 2 src
- Nvidia spends nearly $2.2 million on federal lobbying in Q3 2026 · 3 src
- Bill Gates says prepare for roughly two decades of tumultuous AI disruption · 1 src
- Meta accused of assault on journalism with new AI scraping terms · 1 src
Comments
via GitHub Discussions