Jason Kwon, OpenAI’s chief strategy officer, faced Australian lawmakers in Sydney on Tuesday with an admission that the company’s response to a breach of government websites had fallen short. He said the incident “should not have happened”, that OpenAI “should have handled our response better”, and that it had been a mistake not to contact government ministers directly as soon as it learned of the breaches.
The breach began in June, when an OpenAI agent was researching public medicine spending as part of an internal evaluation. On June 18, after it met access restrictions, the agent found a way to obtain files behind the Medicare Statistics Reporting Service, accessing public and non-public data.
The portal is administered by Services Australia and holds aggregate statistics from Medicare and the Pharmaceutical Benefits Scheme. It is separate from systems that handle individual Medicare claims, payments and personal information. OpenAI said there was no evidence that patient records had been accessed.
Australian officials have described the incursion as one of the most prominent known cases of an AI agent accessing outside systems beyond the United States. Cyber-security experts said it was the first hack of its kind.
The Australian Institute of Health and Welfare said its public-facing website was also among the sites accessed by an OpenAI agent. Officials are working with the Australian Signals Directorate on a forensic investigation into whether other government systems were affected.
OpenAI learned of the breach in August. It notified Services Australia on 10 September through an email sent to a public mailbox, weeks after the June incident.
Kwon told the hearing that OpenAI’s actions had been “not good enough”. “We are sorry and we know we have work to do to rebuild trust with the Australian people,” he said.
OpenAI has since added “more precautions” to its training environments. Its models are now monitored in real time during tests, and an alarm is meant to sound when they interact with the internet in an unintended way.
Kwon said those changes had allowed the company to alert the New South Wales government to another hack last week within 48 hours.
Prime Minister Anthony Albanese condemned the Medicare incident after speaking with OpenAI’s chief executive, Sam Altman, to convey Australia’s extreme concern about both the access and the delay in notifying the government. Albanese said the agent “found a way around those blocks” and “didn’t accept no for an answer.”
Australia has asked Altman and Anthropic’s chief executive, Dario Amodei, in writing to appear before its Senate inquiry into artificial intelligence. Public hearings are due in Canberra on Thursday, 1 October, examining AI’s potential effects on safety, data transparency, Australian communities, industries, water and energy.
Amodei will not attend the Canberra hearing. Dave Orr, Anthropic’s head of safeguards, appeared at the Sydney hearing and said the company had reviewed “hundreds of millions of transcripts” for possible incidents resembling OpenAI’s breach, finding no Australian cases.
The Senate inquiry is scheduled to report by 16 November.
Comments (0)