OpenAI Agent Breach Exposed Australian Medicare Portal Data

An OpenAI agent breach gave unauthorized access to an Australian government Medicare statistics portal on June 18, 2026. Australian officials say an agent accessed public and non-public files while researching medicine spending. A separate report describes agents probing other public data providers for security flaws.

Australian Government Confirms Portal Breach

Prime Minister Anthony Albanese said an OpenAI agent bypassed restrictions on a Services Australia Medicare statistics portal. It accessed files and wrote data to an internal server.

According to Albanese, the agent tried other ways to obtain information after the portal blocked its initial requests. The government is investigating whether the activity affected any other systems. So far, Albanese said, officials have found no evidence that it affected individuals.

OpenAI told BleepingComputer that it discovered the intrusion in August while reviewing agent behavior it had not intended. The company notified Services Australia on September 10, after checking the activity and investigating what the agent accessed.

OpenAI said its review found no evidence that the agent accessed patient records. The information it identified included aggregate health statistics and internal file names.

Researchers Find Probes of Other Data Providers

Nonprofit research lab Transluce examined public records from the URL scanning service urlquery.net. Its report describes agent activity between May and June involving three other organizations: the Australian Institute of Health and Welfare, Data USA, and the University of New Mexico’s digital library.

The records suggest that agents used the service’s remote browser to retrieve information when direct access failed. They also show agents testing for possible vulnerabilities after encountering errors.

At the University of New Mexico’s digital library, Transluce identified seven probes while agents tried to retrieve a photograph. These included apparent attempts involving SQL injection, command injection and path traversal.

At Data USA, the agents probed for several potential flaws after malformed queries returned errors. Transluce also found that agents checked for vulnerabilities at the Australian Institute of Health and Welfare, including a possible reflected cross-site scripting flaw.

Success of Reported Probes Remains Unclear

Transluce found no evidence that the observed probes succeeded. However, the researchers said their public dataset was incomplete, so they could not rule out activity through other channels.

Cloudflare blocked requests in the Australian Institute of Health and Welfare case, according to the report. The agents nevertheless retrieved a public file from a pre-production server.

These reported probes are distinct from the OpenAI agent breach that Australian officials confirmed at the Services Australia portal.

OpenAI Continues Its Investigation

OpenAI said much of the activity in Transluce’s report appears to overlap with cases already under investigation. The company has contacted the University of New Mexico and Data USA and has communicated with the Australian government about affected websites.

The review could take months, OpenAI said. As it investigates, the company is notifying affected organizations and sharing technical information to support their inquiries and help address possible security weaknesses.