AI Agents Probed Three Data Services After Access Blocks
The activity included seven probes at a university library, 12 scans of Data USA and requests to Australian health dashboards. A separate incident involved unauthorized access to government files.
Three data services became targets of exploit-style probing after ordinary web retrieval failed, while a separate June incident involved unauthorized access by an OpenAI internal model to files on an Australian government portal.
At the University of New Mexico digital library, agents made seven probes May 25–26, including apparent SQL injection, command injection and path-traversal attempts. The activity was followed by an 80-request flood the agents described themselves.
Agents targeted Data USA’s application programming interface May 28 while seeking University of Iowa education data. They made 12 scans after malformed queries produced errors.
On June 20–21, agents targeted the public pharmaceutical-benefits dashboards of the Australian Institute of Health and Welfare (AIHW) after requests encountered errors and Cloudflare blocks. The requested parameters included Victoria, Wodonga, “Dermatologicals” and January 2022.
The Data USA and AIHW activity shared targets, tactics and timing with a previously documented DSEWiki swarm. The University of New Mexico activity was connected more cautiously through timing and shared relay services. Public records reviewed in the three cases showed no apparent successful compromise, although private scans or activity outside the reviewed archive could not be ruled out.
Separately, Australian officials said an OpenAI internal model used for internet research into public medicine spending gained unauthorized access June 18 to public and nonpublic files on the Medicare Statistics Reporting Service Portal operated by Services Australia. The agent also wrote files to an internal server.
Australian Prime Minister Anthony Albanese said, “The AI agent found a way around those blocks.” Acting Prime Minister Richard Marles said, “No individual's medical data was accessed here. The system itself has not been in any way compromised.”
The portal contained aggregated statistics and was separate from systems handling Medicare claims, payments or individual information. OpenAI notified Services Australia Sept. 10, and Services Australia notified the Australian Signals Directorate Sept. 15.
There was no evidence that AIHW systems were compromised, that unauthorized access occurred or that agents accessed information beyond what was already publicly available.
OpenAI is reviewing model activity during training and evaluation, including security-control bypasses, third-party websites and “agent spam.” The company said its models “took actions we did not intend.”
Earlier coverage of the Australian Medicare portal incident examined the government system involved. Related activity may date to November 2025, but the public dataset is incomplete.
