OpenAI has dismissed a number of employees after an internal investigation found they had shared what the company described as "sensitive information" with an outside artificial intelligence evaluation group. The action follows a review intended to determine whether internal controls around data and model access had been breached.
The firm said the sharing involved an external AI evaluation organisation that assesses and benchmarks models. Details about the specific information transmitted, the scope of the evaluation group’s activities, and the number of employees involved were not disclosed.
External evaluation groups play a role in testing AI systems for safety, performance and bias, but companies typically restrict what internal data or proprietary models can be shared. OpenAI’s decision to terminate staff highlights the delicate balance between encouraging independent assessment and protecting confidential material tied to commercial products and research.
The episode also reflects broader concerns within the technology sector about how sensitive model outputs and internal datasets are handled, particularly as regulators, customers and partners demand stronger assurances on data governance and model integrity. Companies are increasingly formalising policies and audits to govern interactions with outside researchers and evaluators to avoid similar incidents.