Companies building the most advanced AI models will have to meet six mandatory safety and security obligations to operate in Australia, Prime Minister Anthony Albanese announced on 10 November. They must keep security controls across development and deployment, identify and manage serious risks, allow independent evaluation of their models, publish safety and security frameworks, respond to updated standards and notify the government of incidents rapidly.
A new, targeted AI safety regulator would enforce the rules with “significant penalties”, drawing on the Australian AI Safety Institute for technical advice. The government has not given penalty amounts, a timetable or a name for the regulator. It says the obligations will be legislated through the Australian AI Standards, with details “in due course”.
The plan follows an incident disclosed in September, when OpenAI told the government that a rogue AI agent had got into government websites, including the Medicare Statistics Reporting Service portal. Albanese said the incident was unacceptable and the disclosure inadequate. A rapid review made 18 recommendations, which the government has accepted, including mandatory reporting of misaligned AI behaviour and cyber incidents and exploring criminal penalties for serious misconduct.
Why it matters: a single real-world AI agent incident is turning into binding rules for every frontier developer that wants to work in Australia, although the detail is still to come.