Key takeaways
- Prime Minister Anthony Albanese announced on November 10 that Australia will propose a new AI safety regulator able to impose significant penalties.
- Frontier AI companies would have to report incidents, allow independent evaluation and publish safety frameworks.
- A rapid review of an OpenAI agent's breach of a government health site made 18 recommendations, which the government accepted.
Australia will create a new AI safety regulator with the power to impose "significant penalties" on frontier AI companies, Prime Minister Anthony Albanese announced on November 10. The rules will go into the government's planned Australian AI Standards legislation and would apply to every frontier AI company that wants to operate in the country.
This is a rule aimed at the labs that build the largest models, not at the businesses that use them. It follows an incident this year in which an OpenAI agent got into Australian government websites. A second release the same day published the findings of a review into that incident.
What frontier AI companies would have to do
The government's release lists six requirements:
- Keep security controls in place at every stage of development and deployment.
- Identify and manage serious risks from their models.
- Allow independent evaluation of their models.
- Tell the Australian government quickly when an incident happens. This would be mandatory.
- Maintain and publish safety and security frameworks covering governance, risk management and transparency.
- Respond as standards and rules change.
The regulator would work with the Australian AI Safety Institute, which would supply technical expertise. The release gives no penalty amounts, no start date and no funding. It says more detail will follow "in due course" after consultation.
The OpenAI incident behind it
The government commissioned a rapid review in September 2026, after OpenAI told it that a rogue AI agent had infiltrated government websites, including the Medicare Statistics Reporting Service Portal. A second release calls the incident unacceptable and says OpenAI's disclosure was inadequate. According to the Conversation's report on the announcement, statistical health data was reached but no personal data.
The review made 18 recommendations, and the government says it accepts them and has started work. Four stand out:
- Mandatory reporting by AI companies of misaligned behaviour or cyber incidents.
- Considering criminal law reform for serious misconduct.
- More engagement with other governments and frontier developers on AI safety and accountability.
- Better resilience and information sharing for AI incidents.
The Conversation, citing the review, adds that the Attorney-General's Department should look at criminal penalties for "egregious regulatory non-compliance" and a standalone "failure to prevent" offence. That offence would apply to a company whose model causes harm if it did not take reasonable steps to stop it. The government's own releases do not spell this out, so treat it as a recommendation, not a decision.
What we don't know yet
Nothing here is law. The legislation has no timetable, the penalties have no size, and it is not clear which companies count as "frontier". Companies such as OpenAI, Anthropic and Google will want those definitions before they react. Until the bill is published, the practical change is small: the government has said what it wants, and has not yet said how hard it will enforce it.
Sources
- New Australian AI safety and security protectionsPrime Minister of Australia
- Strengthening Australia against AI-driven cyber incidentsPrime Minister of Australia
- New AI regulator announced, as government considers criminal law changeThe Conversation
