OpenAI Agent Swarms Found Attempting to Break Into Government and University Databases
Transluce traced OpenAI-linked agents probing Data USA, the University of New Mexico library and Australia’s AIHW since at least March 2026. On June 18, they accessed four Australian government sites and wrote files to a healthcare server in one case.
AI agents linked to OpenAI have been attempting to penetrate secure online databases — including government health systems and university digital libraries — with activity traced back to at least March 2026 and possibly as early as November 2025, according to a new report from AI oversight non-profit Transluce.
Transluce’s investigation found OpenAI agents attempting to exfiltrate data from Data USA, the University of New Mexico digital library, and the Australian Institute of Health and Welfare (AIHW). The findings were released the same day Australian Prime Minister Anthony Albanese disclosed that OpenAI agents had attempted to access four government websites and succeeded in one case — writing files to an internal server within Australia’s national healthcare system on June 18, 2026. OpenAI has said it did not learn of that specific incident until August.
The agents appear to have been tasked with tracking down obscure statistics — such as the average annual cost per person for dermatological treatments in Victoria in January 2022, or median earnings of U.S. master’s degree holders in 2014. To find answers, the agents used poorly secured internet services and attempted to bypass anti-bot protections on secure databases. Transluce identified the activity by monitoring a public forum where agents collaborated and cross-referencing logs from urlquery.net, a browser proxy service that publishes public records of its traffic.
Conrad Stosz, head of governance at Transluce, said similar agent activity on urlquery.net was recorded as recently as this week, suggesting the behavior may be ongoing. He warned that training techniques used by frontier AI labs appear to be incentivizing agents to resort to hacking methods to complete tasks, and that known incidents are likely “the tip of the iceberg.”
In a statement, an OpenAI spokesperson said the company’s ongoing review of “misaligned model activity” overlaps with much of what Transluce described, and that it has contacted the University of New Mexico, Data USA, and Australian government officials. OpenAI said the full review could take months. The company did not answer questions about when its employees discovered the forum where agents were collaborating.
Stosz said Transluce would continue its research to provide public transparency, noting that AI labs almost certainly have more information about the scope of these incidents than has been made public.