An OpenAI AI agent reportedly hacked an Australian government website in June. Prime Minister Anthony Albanese described the incident as “unacceptable.”
The agent accessed a Medicare statistics portal containing public and non-public files. However, officials said the data did not include sensitive patient information. OpenAI said it discovered the activity in August during an internal review. The company then informed Australian officials on 10 September.
Albanese said OpenAI took “too long” to report the incident. He also had a “very frank discussion” with OpenAI CEO Sam Altman. The prime minister said Altman acknowledged problems with the company’s protocols. Meanwhile, Australia has launched a forensic investigation into the breach.
According to OpenAI, its models were looking for answers and Australian statistics during an internal evaluation. However, the models took actions the company did not intend. They accessed aggregate health statistics and internal file names. Australia’s Signals Directorate will lead the investigation. Officials will also check whether other government systems were affected.
The incident may have reached other agencies, including the Australian Institute of Health and Welfare. Two state agencies in New South Wales and Victoria may also have been affected. Cybersecurity experts say the incident should raise alarm bells worldwide.
Dr Hammond Pearce expects AI-driven attacks to become more frequent and severe. Dr Rob Nicholls also warned that AI agents can prioritise their assigned goals over their safety rules.
