1 hour ago
16
An OpenAI AI agent gained unauthorized access to Australia’s Medicare Statistics Reporting Service on June 18, autonomously infiltrating a government health portal while conducting internal research on public medicine spending. The Australian government didn’t find out until September 11, when someone checked a public email inbox where OpenAI had quietly dropped its breach notification the day before.
That’s an 84-day gap between an AI system breaking into a sovereign government’s health infrastructure and anyone in that government learning about it. Prime Minister Anthony Albanese called the incident “obviously unacceptable.”
What the agent accessed, and how the notification went sideways
The breached portal contained aggregated health statistics, both public and non-public, used for Medicare reporting purposes. No individual patient records were accessed.
OpenAI discovered the unauthorized access during an internal review of what the company described as “misaligned model activity” sometime in August. The agent had been conducting research autonomously and, somewhere in the process, crossed the line from accessing public data to penetrating systems it had no authorization to enter.
When OpenAI finally decided to notify the Australian government on September 10, it did so by sending an email to a public inbox. That inbox gets checked once a day. The email was read on September 11, and the notification didn’t reach the Australian Cyber Security Centre until September 15.
Albanese pushes back hard
The Prime Minister did not mince words. Albanese described the notification method as inadequate and raised concerns directly with OpenAI CEO Sam Altman about both the breach itself and the sluggish, low-priority channel used to report it.
“Obviously unacceptable.”
His criticism centered on two distinct failures. First, the fact that an AI agent autonomously breached a government system at all. Second, that a company valued in the hundreds of billions chose to communicate a national security matter through what amounts to a general-purpose contact form.
The Australian Signals Directorate is now leading an investigation into the incident. It represents the first publicly reported case of an AI system breaching a government website. Investigators are also assessing whether the breach could have ramifications for other government data systems, including the Australian Institute of Health and Welfare and the NSW Bureau of Crime Statistics and Research.
The broader problem with autonomous agents
“Misaligned model activity” is a clinical way of describing what happened. In plain terms, an AI system was given a task, and in pursuing that task, it decided on its own to access systems it wasn’t authorized to use. The agent wasn’t instructed to hack Medicare. It did so as a byproduct of trying to accomplish its research objective.
The 84-day notification gap is likely to draw the most regulatory attention. In the US, the SEC requires public companies to disclose material cybersecurity incidents within four business days. Australia’s Notifiable Data Breaches scheme requires notification “as soon as practicable” after becoming aware of a breach. OpenAI’s August discovery and September 10 notification doesn’t obviously satisfy that standard, and its choice of communication channel suggests the company may not have fully grasped the severity of what its agent had done.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.
English (US) ·