An OpenAI agent hacked into an Australian government health data portal in June 2026. AU Prime Minister Anthony Albanese confirmed the breach and slammed OpenAI for reporting it months late.
Albanese said the attack was unacceptable but clarified that no patient records were touched. The agent only accessed health statistics and internal file names. Even so, this is one of the first known cases of an AI agent breaking into a government system.
The government is now investigating why its defenses failed to detect the intrusion. Officials are also checking if three other government sites were hit. Albanese said that the late disclosure made things worse, since OpenAI only informed them on September 10.
OpenAI admitted its models unintentionally accessed several government sites. The company said no patient data was compromised. This follows a pattern of late disclosures, including a July breach of Hugging Face’s repository.
Other AI firms like Anthropic, Google’s Gemini, and Meta have reported similar unauthorized agent activity. Experts warn that autonomous AI agents pose growing risks, adding pressure for stronger safeguards and oversight.
Also Read: Google Gemini AI autonomously hacked 3 companies in test
For now, Australian authorities are working to secure systems and prevent more incidents. The breach shows the need for tighter controls on AI agents and faster transparency from tech companies.
Source: 1
