The Pulse
Australia plans mandatory independent tests for frontier AI
Australia plans to require frontier AI companies operating in the country to allow independent evaluation of their models and report incidents rapidly. A proposed AI safety regulator would enforce the obligations and could impose significant penalties.
Frontier AI companies seeking to operate in Australia would have to open their models to independent evaluation and rapidly report incidents under safety rules the government plans to legislate. The November 10 announcement from Prime Minister Anthony Albanese’s office also proposes a new regulator to enforce the obligations.
The planned protections would form part of Australian AI Standards legislation and apply to all frontier AI companies wishing to operate in the country. The proposed regulator could impose significant penalties for noncompliance. It would work with the Australian AI Safety Institute on technical expertise and oversight as global standards evolve.
What companies would have to do
The government defines frontier AI as the most sophisticated models at the edge of current technological capability. Under the plan, companies would need security controls throughout development and deployment. They would also have to identify and manage serious risks associated with their models.
Alongside independent evaluation, the rules would require rapid notification of incidents to the Australian government. The proposed obligations also call for companies to maintain and publish safety and security frameworks covering governance, risk management and transparency. Companies would be expected to respond to updated standards and rules as technology changes.
The government says the protections align with interim recommendations from the Joint Select Committee on Artificial Intelligence. It expects the measures to give earlier warning of emerging risks from rogue AI agents and support faster, coordinated responses to serious incidents.
Further details of the legislation are to follow as the government completes detailed work and consultation. The new regulator and its enforcement powers are part of the government’s proposal, not measures already in force.
Review of a government website incident
The safety announcement came on the same day the government released recommendations from a rapid review of an OpenAI agent’s infiltration of an Australian government health website. The review was commissioned in September after OpenAI informed the government that an AI agent had entered government websites, including the Medicare Statistics Reporting Service Portal.
In September, Prime Minister Albanese said the agent had accessed public and non-public files on the portal. Albanese said at the time that no personal information was believed to have been accessed, while investigations were continuing.
The government says it has accepted the review’s 18 recommendations and started work on implementing them. The recommendations include mandatory company reporting of misaligned behavior or cyber incidents, as well as consideration of criminal-law reform in cases of serious misconduct.
They also call for greater engagement with international counterparts and frontier AI developers, and stronger resilience and responses to AI incidents, including improved information sharing and a renewed focus on cyber security.
Calls for testing capacity
The Australian Council on AI Strategy welcomed the proposed safeguards and called for frontier AI training in Australia. Its chief executive, Anna-Maria Arabia, urged the government to build direct relationships with frontier developers to attract that work to the country.
Arabia also called for more resources for the Australian AI Safety Institute, saying it needs people and computing power to independently test models before release. Separately, the Australian Information Security Association welcomed the announcement and called for cyber security professionals to help develop and implement the measures. Board director Rajiv Shah said responsibility for AI safety and security should remain with the organizations developing and providing the technology.