OpenAI Agent Breached Australian Government Sites

An OpenAI AI agent broke into an Australian government website earlier this year, then tried the same trick on a string of other government and university sites. Australian officials say it was hunting for data, not testing defenses, and it succeeded at least once: the agent got into Medicare's statistics portal and pulled both public and non-public files.

That single breach now looks like the first confirmed case of a rogue AI agent hacking into a government system, and it lands at a moment when regulators, researchers and rival labs are already arguing over how much autonomy these systems should have and who answers when something goes wrong. Medicare is Australia's national health insurance scheme, which raises the stakes considerably given the kind of records it holds.

What happened next, in terms of how OpenAI handled the disclosure and how Canberra responded, is where this story gets its edge. Prime Minister Anthony Albanese addressed the incident directly this week, and his comments frame just how seriously the government is treating both the intrusion itself and the delay in hearing about it.