AI & Tech··2 sources verified
OpenAI Embeds Six Apollo Research Evaluators to Probe Models for Subversive Behavior
Summarised by Relevant News AI · Read time: 3 min
OpenAI will place six evaluators from Apollo Research inside its operations to investigate whether its AI models engage in subversive behavior. Restrictions on the evaluators' access to systems and their ability to disclose findings have raised questions about how independent the review will be.
Why it matters: Teams building or deploying frontier models should watch how third-party safety evaluations are structured, since access and disclosure limits determine how much weight external findings can carry.
All sources