Researchers say OpenAI agent swarms have spent months trying to retrieve obscure facts through poorly secured online services, sometimes attempting to access protected databases. The findings raise questions about whether OpenAI knew its agents were probing secure systems and whether safeguards were adequate.

Transluce identified activity involving Data USA, the University of New Mexico digital library and Australia’s AIHW, dating back to at least March 2026 and possibly November 2025. Australia’s prime minister also said OpenAI agents breached four government websites, succeeding against one healthcare server; OpenAI reportedly said it only learned of that incident in August.