External auditors from METR and Redwood Research have completed a three-week on-site audit at OpenAI, revealing serious gaps in the company’s internal monitoring system for AI agents. The auditors found that up to 25 percent of agent traffic from models classified as GPT-5.5 and above was not being monitored, raising concerns about the safety and oversight of these advanced AI systems.
The audit report, which spans 61 pages, has been largely redacted, with only a summary made publicly available. This lack of transparency has led to increased scrutiny regarding OpenAI’s commitment to ensuring the safety of its AI technologies. The findings underscore the ongoing challenges in effectively monitoring advanced AI systems as they continue to evolve.
Compiled automatically by the Tech AI Newsdesk from public AI-news sources and summarised in our own words.
Frequently asked questions
What did the audit reveal about OpenAI's monitoring system?
The audit revealed serious gaps, with up to 25 percent of agent traffic from models classified as GPT-5.5 and above not being monitored.
Why is the audit report's redaction significant?
The redaction of the audit report has led to increased scrutiny regarding OpenAI's commitment to ensuring the safety of its AI technologies.
What are the implications of the audit findings?
The findings underscore the ongoing challenges in effectively monitoring advanced AI systems as they continue to evolve.
