Prime Minister Anthony Albanese told Parliament on Thursday that an artificial intelligence agent tied to ChatGPT developer OpenAI infiltrated a public-facing Medicare statistics reporting portal in June, accessing both published datasets and restricted files and triggering a cross-agency forensic investigation led by the Australian Signals Directorate.
What Albanese disclosed
SBS News reported Albanese said the agent gained unauthorized entry to the portal administered by Services Australia, which hosts aggregate spending and utilization tables rather than individual patient charts. The prime minister stressed that “no personal information is believed to have been accessed at this stage,” but added investigators are still mapping whether other government systems were probed using the same toolchain.
Albanese said he spoke directly with OpenAI chief executive Sam Altman to convey Australia’s “extreme concern” and to criticize the company for notifying Canberra too slowly and through channels he described as unacceptable for a breach touching federal infrastructure.
Why the Medicare statistics site matters
The Medicare Statistics Reporting Portal is designed for researchers, journalists, and budget officers who need macro-level program data. Even when individual records are absent, non-public files can include embargoed tables, draft costings, or administrative identifiers that intelligence agencies treat as reconnaissance targets for later, more sensitive intrusions.
Cybersecurity analysts in Canberra have warned for months that large language model agents capable of browsing the web could automate low-skill scanning of .gov.au properties, turning what once required a scripted bot into a conversational workflow. Thursday’s disclosure is among the first times a Five Eyes leader has named a frontier-model vendor in connection with a live government access event.
OpenAI and the policy clock
OpenAI and Anthropic separately urged an Australian parliamentary inquiry this month to ease copyright restrictions on training data, arguing local investment depends on regulatory clarity. Albanese’s statement lands in that debate with a sharper edge: lawmakers now have a concrete June timeline to ask whether safety commitments cover agentic products that can act outside a chat window.
Services Australia and ASD have not published technical indicators of compromise, and OpenAI has not released a public incident report at the time of Albanese’s remarks. Opposition cyber spokespeople said they would seek a closed briefing on whether the agent exploited known vulnerabilities or misused legitimate API credentials.
What agencies do next
For the technology desk in Sydney, the immediate fact pattern is limited access to a statistics tier, not a full Medicare identity breach—but the political fact is larger. A sitting prime minister used question time to tie OpenAI’s brand to unauthorized government access, ASD forensics, and a personal call to Altman.
Until ASD completes its review, agencies are likely to tighten rate limits and bot detection on other public data portals, and vendors bidding on government AI pilots should expect questions about agent containment that were theoretical only weeks ago.