OpenAI is bringing four independent safety evaluators inside its systems as part of a pilot to strengthen model safety and oversight. Across outlets, the evaluators are described as assessing aspects of AI behaviour and operational safeguards, including model alignment, protections against unsafe outputs, and ongoing monitoring.
The pilot is also framed as increasing transparency and external scrutiny. Quartz reports the evaluators receive access to OpenAI’s code and training data as part of their assessments. CNBC adds context that independent evaluation has become more prominent after a series of public AI safety and security incidents, which have heightened attention on how organizations test and validate their systems.
While both outlets focus on OpenAI’s partnership approach, Quartz emphasizes the technical scope of the evaluators’ work and the level of access, whereas CNBC highlights the broader backdrop and why such external checks are gaining importance. Neither outlet, based on the provided materials, specifies timelines, evaluator identities, or immediate outcomes from the pilot.