OpenAI apologizes for access to Australian government systems

After Australian lawmakers called company executives to an inquiry over an AI agent's access to a Medicare portal, OpenAI apologized in a company post for the incident and acknowledged that it should have notified affected agencies sooner.

During internal training in June, an experimental model gained nonpublic access to a Services Australia service. It ran commands, retrieved internal files, credentials and aggregate statistics, and wrote files. OpenAI said its review found no evidence that the model accessed individual medical or client records.

OpenAI also identified activity involving systems run by the NSW Bureau of Crime Statistics and Research, the Victorian Department of Health, and the Australian Institute of Health and Welfare. The company said no individual crime or medical records were accessed in those cases.

The company now blocks live internet access in relevant research environments, uses cached web content and has expanded monitoring. It has paused training and evaluation involving tool use for its most capable models. OpenAI will also establish an Australian taskforce, provide technical support and funding for cyber defences, and send Chief Strategy Officer Jason Kwon to appear before the Joint Select Committee on Artificial Intelligence in Sydney on October 6.