OpenAI opens its AI systems to four independent on-site auditors for the first time

What happened
OpenAI has allowed four independent audit groups to access its internal AI systems on-site for the first time. The auditors are tasked with assessing safety risks and model behavior, as part of commitments made under the White House Accord on Super Intelligence. The groups will review unreleased models, source code, training data, and monitoring systems at OpenAI's offices. Their work is overseen by an independent safety board, with plans for ongoing and follow-up reviews.
Key details
- The four audit organizations are Apollo Research, METR, Transluce, and Goodfire.
- Auditors can speak confidentially with OpenAI developers during their reviews.
- Apollo Research is investigating if models act against developer intentions.
- METR is reviewing behavior in the largest reinforcement learning training run.
- OpenAI and Transluce plan an eight-week follow-up review in early 2027.
Why it matters
External auditors, AI researchers, and OpenAI staff are directly affected, as this process gives outside experts unprecedented access to internal systems. The audits may lead to changes in how OpenAI develops and monitors its models, and could result in fixes or adjustments before public disclosure of findings. The independent safety board’s oversight and the ability for auditors to publish their findings, with some redactions or delays, introduce new accountability measures for OpenAI’s AI development.
- OpenAI inks agreements with four independent evaluators to boost company's safety work· 2 sources
- OpenAI will embed safety researchers to investigate AI subversion· 3 sources
- Building trustworthy AI by fencing it in, not predicting it· MIT Technology Review
- Our first phase of embedded assessments· Open AI News
- The Biggest AI Agent Risk Is The Workaround Nobody Approved· Forbes
- OpenAI agents keep disabling monitoring; is this being handled appropriately?· LessWrong
