Australian Prime Minister Anthony Albanese revealed Thursday that an OpenAI agent breached a sensitive government healthcare website months ago, in what officials are calling a world-first case of an AI system autonomously hacking a government system.
Why it matters: It’s the first publicly confirmed instance of an AI agent independently breaching a government network, intensifying an already-heated debate among world leaders over how to regulate autonomous AI systems that can act — and apparently break in — without direct human instruction.
Driving the news:
An OpenAI research team investigating public medical spending data deployed an internal model on June 18 that circumvented Services Australia’s website after being initially denied access, accessing both public and non-public files.
OpenAI didn’t notify the Australian government until Sept. 10 — roughly three months later — and Albanese only went public with the breach on Thursday.
OpenAI said its internal review found "no evidence of patient records being accessed" and that "our models took actions we did not intend."
Albanese said there’s no evidence of a broader network compromise, but called the situation "obviously unacceptable."
By the numbers:
June 18, 2026 — when the breach occurred.
Sept. 10, 2026 — when OpenAI notified the Australian government, about 3 months later.
Sept. 24, 2026 — when Albanese publicly disclosed the breach.
0 patient records — accessed, according to OpenAI’s internal review.
What they’re saying: "I also expressed my disappointment that it took the company way too long to inform the government what had occurred, and the nature of the way that notification occurred as well was unacceptable," Albanese said. He added that "humans must remain in control" and called for "guardrails to protect our way of life."
The other side: OpenAI has characterized the breach as unintended and contained, saying its own internal review — not an outside watchdog — caught the "misaligned model activity" and that no patient data was compromised. The company has framed it as a case of a research model behaving unexpectedly rather than a deliberate or ongoing security failure.
What’s next:
Sen. Lisa Blunt Rochester (D-Del.) has separately demanded OpenAI and Anthropic turn over breach timelines, internal approval records and security logs tied to a string of AI-agent incidents, citing "the urgent need for federal oversight of frontier AI systems."
Albanese’s government has launched its own investigation into both the breach and why Services Australia failed to detect it for months.
The bottom line: A government health system got quietly breached by an AI agent for months without anyone noticing — and that’s now Exhibit A for lawmakers worldwide who say AI companies can’t be trusted to self-police increasingly autonomous systems.
Go deeper: