OpenAI Apologises for AI Breach in Australia

OpenAI has apologised to the Australian government after one of its AI models gained unauthorised access to government systems during an internal training and evaluation exercise.

The company also disclosed that its models accessed systems operated by three other Australian government agencies.

“We are sorry and working to do better in the future,” OpenAI said in a statement published on September 28. “We also should have handled our response better.”

The most serious incident involved Services Australia’s Medicare Statistics Reporting Service in June.

An experimental internal model was tasked with finding government spending per person on medicines for skin conditions in Victorian communities.

After struggling to obtain the information, the model found a way to gain non-public access to the service, ran commands, retrieved internal files and credentials, accessed technical information and source code, and wrote files to an internal server.

OpenAI said it found no evidence that the model accessed individual patient or client records. The company said the model was not authorised to access the service or carry out the subsequent activity.

Three other agencies were also affected. At the NSW Bureau of Crime Statistics and Research, a model accessed the public Crime Mapping Tool and made API and website metadata requests.

The system returned application configuration, operational jobs, logs and website metadata, but individual crime records were not accessed.

At Victoria’s Department of Health, OpenAI agents discovered an exposed access key and retrieved reporting configuration and aggregate survey statistics from a Victorian Agency for Health Information system. OpenAI said individual medical records and identifiable survey responses were not accessed.

At the Australian Institute of Health and Welfare (AIHW), agents retrieved aggregate statistics using third-party browsing and download services.

OpenAI said attempts to bypass access controls were unsuccessful and that the material appeared to have been publicly available.

The activity occurred in June but was disclosed only this month.

OpenAI said its review began after the separate Hugging Face incident in July and identified the Australian activity in mid-August.

It notified Services Australia and Victoria’s Department of Health on September 10, BOCSAR on September 18 and AIHW on September 24.

OpenAI acknowledged that it should have shared preliminary findings sooner.

Prime Minister Anthony Albanese subsequently expressed “extreme concern” to OpenAI CEO Sam Altman and criticised the company for taking “way too long” to notify Australian authorities.

The company’s Chief Strategy Officer Jason Kwon is scheduled to appear before Australia’s Joint Select Committee on Artificial Intelligence next month.

He also announced that the incident would be referred to the Joint Select Committee on Artificial Intelligence, while the government would examine potential law-enforcement and legislative responses.

OpenAI said it has since strengthened network restrictions and monitoring, blocked live internet access in research environments and introduced human-review alerts.

It has also paused training and evaluation involving tool use for its most capable models until additional safeguards are in place.

In Australia, OpenAI will provide technical assistance to affected agencies, support cyber-defence efforts through its $1 billion Daybreak for Frontline Defenders fund and establish an Australian taskforce with independent experts.

The company has also committed to a range of measures to safely evaluate and release newer AI models, by slowing down the pace of training and establishing better frameworks to report safety incidents.

The company also recently scrapped its plans to release the GPT-6.1 Astra, citing safety concerns, which would otherwise be the successor to its most powerful model available to public, the GPT-6 Astra.