Australia’s government has announced legislation that would impose mandatory safety duties on companies operating the most advanced, high-risk AI models in the country. The proposal sits within its “AI in Australia’s Interests” agenda and would replace voluntary assurances with enforceable requirements for the covered developers.
Companies would have to permit independent model audits, maintain ongoing safety checks, publish their safety frameworks and report security incidents to the government immediately. A new oversight body would work with the Australian AI Safety Institute and could impose substantial penalties for violations.
The incident-reporting provision is intended in part to surface warning signs from autonomous agents before failures spread. The government says it wants domestic guardrails in place before entering broader international negotiations, while creating enough regulatory certainty for investment and uses in areas such as healthcare.
Important boundaries are not yet settled. The announcement does not fully define which models cross the frontier-risk threshold, what technical access auditors will receive or how penalties will be calculated. Those details are expected after consultation. The measure is therefore a policy commitment rather than an operating audit regime today, but it signals that developers serving Australia may face recurring external testing instead of one-time self-assessments.