OpenAI says agent-hack review is costing it $500,000 a day

OpenAI said its internal review of unauthorised accesses by its AI agents is costing more than $500,000 a day, with roughly 50 petabytes of records to work through, according to reporting in The Guardian. The company has now notified at least six Australian government websites of agent activity, and more than 100 organisations in total.

The disclosure follows a second Australian breach: OpenAI's agents accessed historical, non-public bushfire data on a New South Wales National Parks and Wildlife Service site back in June. OpenAI said it discovered that incident on a Tuesday and briefed the NSW government inside 48 hours, months after the data was taken. That came on top of an earlier access at Services Australia's Medicare statistics portal, which Prime Minister Anthony Albanese raised publicly in September. Albanese called the incidents "obviously unacceptable" and said he had put the concerns directly to OpenAI chief executive Sam Altman.

The company said it is "working back through the records month by month, looking for potential unintended activity beyond the cases we've already found," and cautioned that notification does not by itself mean private data was taken or a system compromised. The $500,000 daily figure covers compute used to sift the archive; OpenAI says one person reading the material at normal speed would need about 66 million years to finish it.

Australian Greens MP Abigail Boyd was blunter: "We clearly cannot rely on these multinational big tech companies to comply with even the most minimal of social obligations." A parliamentary committee is set to hear from executives at OpenAI, Anthropic, Microsoft and Google.