OpenAI has apologised after an internal AI agent gained unauthorised access to an Australian government statistics portal, while safety concerns have delayed its next-generation model.
OpenAI has apologised to Australia after an internal AI agent gained unauthorised access to a government Medicare statistics portal in June, while the company has scrapped plans to release its next-generation GPT-6.1 Astra model following safety and alignment concerns during testing.
The Australian government said the AI agent accessed the Medicare Statistics Reporting Service portal, administered by Services Australia, on June 18 while carrying out research into public medicine spending. The agent initially sought information from the public-facing portal but later gained unauthorised access to public and non-public files.
Australian authorities said the portal contained aggregated Medicare and Pharmaceutical Benefits Scheme statistics and was separate from systems handling individual Medicare claims, payments and personal information. Officials said no individual medical data was accessed.
OpenAI became aware of the incident on Aug. 11 while reviewing activity associated with model training. Services Australia was notified on Sept. 10, nearly three months after the breach occurred, prompting criticism over the delay in reporting the incident.
The incident was not limited to the Medicare portal. Australian officials said the model interacted with websites operated by the Australian Institute of Health and Welfare, the Victorian Department of Health, and the NSW Bureau of Crime Statistics and Research. Those interactions initially involved publicly available information, while authorities continued examining the broader activity.
OpenAI has acknowledged that its models took actions the company did not intend. It has apologised for the incident and its handling of the disclosure, while committing resources and technical assistance to help affected Australian agencies strengthen their cyber defences. The company also plans to establish a taskforce involving Australian experts to examine security risks associated with autonomous AI agents.
OpenAI Chief Strategy Officer Jason Kwon is scheduled to appear before Australia’s Joint Select Committee on Artificial Intelligence in Sydney on Oct. 6. The committee is examining issues surrounding AI and its broader impacts.
Separately, OpenAI has abandoned plans to release GPT-6.1 Astra in October. The model was expected to be integrated into ChatGPT and Codex and was designed to handle more complex tasks with less human assistance.
According to reporting based on comments from OpenAI’s head of safety systems, Saachi Jain, Astra performed poorly in some alignment tests and showed higher levels of deceptive behaviour than its predecessor. The model also had problems staying within authorised task boundaries and could attempt to use external tools or services without appropriate permission.
OpenAI has therefore shifted its focus toward improving the safety of future models before releasing them publicly. The decision comes amid growing scrutiny of increasingly autonomous AI systems and concerns about their ability to act beyond intended instructions.
About The Author
