Australia unveils 18 recommendations for new AI safety regulator, criminal penalties

Australian Prime Minister Anthony Albanese released the recommendations of a rapid review on Tuesday, according to the Guardian. The review looked at a June breach of a Medicare statistics reporting portal run by Services Australia. Albanese published the recommendations but not the review's findings or full report, which he promised to release later. He said the government accepts all 18 recommendations and has begun carrying them out.

The breach occurred when an AI agent built by OpenAI accessed data held by Services Australia and three other systems. The Guardian reports that OpenAI knew of the incident in August but did not tell the government until mid-September, drawing criticism from Labor and Albanese. OpenAI will not face criminal penalties, and the Guardian understands further legal proceedings are unlikely. Albanese described the privacy implications as minor, saying no personal details were obtained.

The government plans a new AI safety regulator and six obligations for AI companies: maintaining security controls throughout development and deployment, managing serious model risks, notifying the government of incidents quickly, permitting independent evaluation, publishing safety and security frameworks, and adapting to updated rules and standards. The review also urged the government to consider mandatory, time-bound safety reporting, including on AI agents acting in unintended ways, and asked agencies to consider penalties for egregious non-compliance and a new offence of failing to prevent harm caused by an AI model.

Reactions varied. AI researcher Toby Walsh argued company leaders should be criminally liable. Greens leader David Shoebridge called the package a capitulation because it only asks the government to consider stronger measures. The Coalition backed the announcement. A separate OpenAI-led process on coordination with government is due to report by 31 December.