An Australian parliamentary committee has advised Prime Minister Anthony Albanese to require independent safety checks on artificial intelligence companies and to give security agencies such as the Australian Signals Directorate early access to new models as a condition of frontier labs training them in Australia, ABC News reported. The proposals would put embedded evaluators and mandatory incident reporting on the table in a close U.S. ally, and any American AI developer that wants to build or train models in Australia would have to plan for them.

What Happened

The Joint Select Committee on Artificial Intelligence, chaired by Labor lawmaker Jo Briskey with Liberal Senator Jonno Duniam as deputy, sent Albanese a letter of “early advice,” the Australian Broadcasting Corp. reported.

According to the ABC, the committee recommends placing independent experts inside leading AI companies, with access to their systems and a duty to report safety findings and incidents to government, and requiring frontier labs to report incidents in a timely manner. The Australian Signals Directorate is the country’s signals intelligence and cybersecurity agency.

RelatedWhite House AI Task Force Plans Breach Warnings From Labs at Its First Meeting

Other recommendations include reserving a share of computing capacity for national security and university research, an urgent review of protections for government data repositories, especially those holding sensitive or national security information, monitoring the risk that AI helps develop biological weapons, a review of copyright settings, and national standards for data centers on resource use, extra electricity generation and community consultation. The committee also wants a standing parliamentary committee on AI that would receive regular reports on AI incidents.

The advice has drawn a process dispute. Independent Senator David Pocock said an adviser to Albanese “pre-vetted” the letter and that the prime minister’s office marked up a draft. Briskey said she had engaged with the office to seek advice and coordination and that the final report was not predetermined. “There has been no guidance, no interference,” Duniam told the ABC.

Why It Matters

Australia is one of the Five Eyes intelligence partners, with the United States, Britain, Canada and New Zealand. A requirement that its signals agency see models before release would give a foreign security service a formal look at U.S. companies’ systems, which labs would need to weigh against their own security policies.

The proposals also track a wider pattern. Embedded evaluators with reporting duties, mandatory incident disclosure and a parliamentary body that receives incident reports would add Australia to a growing list of governments asking frontier developers to show their work. Britain said on Oct. 12 it would examine AI incident and near-miss reporting under a pending cyber bill.

Context

The inquiry follows revelations last month that OpenAI’s experimental models accessed Australian government websites without authorization, including a Services Australia portal hosting Medicare statistics. The government opened an investigation, and OpenAI apologized. OpenAI chief strategy officer Jason Kwon has said the company would support a framework of mandatory disclosures, the ABC reported.

What to Watch

The committee’s final report is due in November, and Albanese will decide which recommendations to adopt. The government hopes to introduce legislation on national standards for data centers and frontier AI models by the end of 2026 and pass it in early 2027. Watch whether the early-access condition survives into the bill and how labs with Australian expansion plans respond.