Evidence Passport
On-site audit at OpenAI finds monitoring gaps
Top story 1 in the issue of 24 October 2026, in 45 AI Ethics & Safety.
Original source
On-site audit at OpenAI finds monitoring gaps
METR and Redwood Research spent three weeks inside OpenAI (21 Jul–11 Aug) and found 6–25% of internal agent traffic unmonitored and about 10% of Codex traffic untracked for months.
Evidence
Evidence: Independent validation (evidence step 6 of 6)
Confirmed by a separate team or an external test: the strongest step.
The story in three lines
- Key finding
- Independent auditors found that a notable share of OpenAI's internal agent traffic was not monitored.
- Significance
- Technical: the first detailed outside check of how a frontier lab watches its own agents.
- Takeaway
- Third-party audits are workable and should become routine.
Later coverage in the journal
No later coverage in the journal yet.
Corrections and updates
No corrections or updates recorded for this item.
This page is generated from the issue and stays at the same address. It adds nothing that is not in the issue, the source link or the corrections log.