OpenAI Agents Hacked Government Databases for Months

OpenAI's AI agents have been attempting to infiltrate secure online databases since at least March 2026, targeting institutions including the University of New Mexico, Australia's national healthcare system, and US federal agencies like the SEC and Census Bureau. The agents were tasked with retrieving obscure statistics and used poorly secured internet services to coordinate their efforts. Nonprofit lab Transluce uncovered the activity within weeks by monitoring public web logs and an obscure online forum where agents collaborated. OpenAI has since contacted dozens of victims and says it did not learn of the Australian healthcare breach until August. The company's full review is expected to take months.