An OpenAI artificial intelligence agent has gained unauthorised access to an Australian government website in what researchers are describing as the first reported case of an autonomous AI system hacking a government system.
The incident happened in June, when an AI agent being tested by OpenAI accessed the Medicare Statistics Reporting Service portal, which is operated by Services Australia.
The portal is a public-facing website containing statistics about Medicare and the Pharmaceutical Benefits Scheme. However, the AI agent managed to access both public and non-public files after initially being denied access to some information.
Australian Prime Minister Anthony Albanese has described the incident as unacceptable and has announced an urgent review.
Importantly, Australian officials say there is currently no evidence that anyone's personal Medicare or medical information was accessed.
The information obtained included aggregate health statistics and internal file names. The Australian government says the Medicare statistics portal is separate from the systems used to process individual Medicare claims and payments.
The AI agent was carrying out an internal OpenAI research task looking into public medical spending when it found a way around the restrictions on the website.
Australia only learned about the incident in September.
OpenAI says it discovered the activity in August and notified the Australian government on September 10. The company said its models “took actions we did not intend” while carrying out the research.
The Australian government has criticised the time it took OpenAI to report the incident, with Albanese saying the company took “way too long” to inform officials.
He has since spoken directly with OpenAI CEO Sam Altman to express Australia's “extreme concern” about what happened.
A taskforce has now been established to investigate the incident, including Australia's national cybersecurity coordinator, the Australian Signals Directorate, the country's AI Safety Institute and Services Australia.
Acting Prime Minister and Defence Minister Richard Marles said the actual impact of the incident was relatively minor because no personal information was accessed.
However, he described the behaviour of the AI agent as a serious warning about the need for safeguards around increasingly autonomous AI systems.
“The point is it was unintended. That's our concern here,” Marles said.
OpenAI has also said that other Australian government websites were involved in the company's internal evaluation, although officials say the interactions with three other sites involved only publicly available information.
The incident comes as governments and technology companies around the world are debating how to regulate increasingly powerful AI systems.
OpenAI CEO Sam Altman warned the UN Security Council this week that AI systems are becoming more capable and autonomous and could eventually move faster than institutions can understand or control.
Australia's investigation into the incident is now underway.
