Australia proposes binding safety duties for makers of top-tier AI models

Australia's government wants frontier AI developers to accept independent audits, publish safety frameworks and report security incidents right away.

Australia's government, led by Prime Minister Anthony Albanese, has announced a proposed law that would put mandatory safety rules on frontier AI companies operating in the country. It would be the first time Australia has imposed rules of this kind. They would apply to the most advanced, highest-risk models, and sit under the government's 'AI in Australia's Interests' agenda.

Developers would have to allow independent audits of their models, keep up ongoing safety checks and publish their safety frameworks. They would also have to report security incidents to the government immediately, a requirement partly meant to catch early signs of AI agents behaving erratically. A new oversight body would work with the Australian AI Safety Institute to enforce compliance and issue large penalties for violations. Albanese, Deputy PM Richard Marles and ministers Tony Burke and Tim Ayres presented the plan.

For security, this moves AI risk from voluntary promises toward outside checking. Independent audits and published frameworks mean the public and regulators can see how the most capable systems are tested, instead of relying on each company's word. The government says it is setting domestic guardrails before entering international negotiations, so the design could shape how other countries approach the same problem. It also says the aim is trust and regulatory stability that attract investment and let AI benefit the economy and healthcare. That touches knowledge too, since the systems in question are widely used tools.

What to watch: this is a proposal, and further details are due after consultations conclude. No penalty amounts, timeline or definition of 'frontier' model have been given yet, and those will decide how much the rules actually bind.

Source: The Decoder