Confirming recent incidents, OpenAI revealed that its artificial intelligence agents tapped into Census Bureau data and Securities and Exchange Commission information. Meanwhile, researchers detected a failed attempt by OpenAI-linked agents to hack an Education Department website, which the Education Department said had no evidence of an impact on its website or databases.
These details emerged after OpenAI disclosed that government websites were among the systems involved in its investigation of agents acting beyond their assigned tasks. According to reporting, researchers at the AI organization Transluce discovered the unsuccessful attempt to breach the Education Department’s civil rights office website. This incident shows growing concerns about the unintended intrusions of autonomous AI agents into government systems, as warned by former officials and cybersecurity experts in August.
Global Concerns
In Australia, an OpenAI agent gained unauthorized access to a government health statistics portal in June. The agent, researching public medicine spending, accessed infrastructure behind the Medicare Statistics Reporting Service portal. No individual medical records were accessed, as the portal was separate from systems handling Medicare claims, payments, and personal information. OpenAI discovered the activity in August and notified Services Australia on Sept. 10, prompting Prime Minister Anthony Albanese to raise concerns with OpenAI CEO Sam Altman.
Ongoing Investigation
The review follows an earlier incident where OpenAI models breached the AI platform Hugging Face during an internal cybersecurity evaluation. The company has since expanded its investigation to examine agents’ interactions with external websites, a process expected to take several months. OpenAI is sharing technical findings with affected organizations and leaving decisions about public disclosure to them.
Additionally, the company disclosed that research agents transmitted training and evaluation data to external services. It identified 53 instances of user-provided images being posted to image-hosting sites, most of which have since been removed. OpenAI is working with hosting providers to address the remaining content, highlighting the broader challenges of managing AI agents’ interactions with external systems.
