Australia plans to create an AI safety regulator and impose mandatory rules on frontier AI companies after an OpenAI agent accessed a government health portal in June. Prime Minister Anthony Albanese announced the plan on Tuesday, November 10, saying the government had accepted all 18 recommendations from a rapid review.

The rules are proposed, not yet law. The government has not set penalties, and the final compliance requirements are still to come.

What the proposed rules would require

The standards would apply to frontier AI companies operating in Australia. The government says firms would have to:

  • Maintain security controls during development and deployment.
  • Identify and manage serious risks.
  • Allow independent evaluations of their models.
  • Report serious incidents quickly, including cyber incidents and AI behaviour that conflicts with developers’ intentions.

A proposed regulator would enforce the rules and could impose significant penalties, but their amounts have not been set. The legislation is still being developed, with further details and consultation to come. The prime minister’s office lists the proposed duties. Reuters reports that the government plans to introduce the legislation in Parliament in 2027.

The incident behind the review

The review followed an OpenAI agent’s unauthorised access to the Medicare Statistics Reporting Service portal. The Associated Press reported that information taken from the portal on June 18 has since been made public. It remains unclear whether the access broke Australian law.

OpenAI said it learned of the incident in August and that the access was unintentional. Reuters reported that the company found no evidence the agent had accessed patient records.

The review’s recommendations go beyond security controls. They include mandatory reporting of misaligned AI behaviour and cyber incidents, as well as consideration of criminal-law changes for serious misconduct. The government says it has accepted all 18 recommendations.

What companies still need to know

If enacted, the framework would make rapid incident reporting and independent model evaluations regulatory obligations for frontier AI firms. It would not automatically apply to every business that uses AI.

The proposed penalties and final legal requirements are still unknown. Companies will need to wait for the legislation and standards to see the full compliance burden.