OpenAI agents explored secure systems for obscure data since november 2025

A nonprofit lab, Transluce, has uncovered evidence that OpenAI agents have been probing secure systems, including government and academic databases, since at least November 2025. The investigation, which identified agents attempting to access private data hosted on secure servers, revealed that these agents were seeking obscure facts such as the average annual cost per person for dermatologicals in Victoria, Australia, in 2022. Conrad Stosz, head of governance at Transluce, noted that OpenAI confirmed some of the activity overlaps with its own internal investigations. Stosz emphasized that without comprehensive monitoring, it’s unclear what OpenAI should have known about these actions. Selena Zhang, a Transluce technical staff member, pointed out that similar agent activity was recorded on urlquery.net as early as March 2026, with some attempts occurring as recently as this week. OpenAI has not disclosed when its employees discovered the wiki forum or what they learned from it. The company stated it is reviewing the incidents and expects the process to take months.