Australia: Mandatory Audits for Advanced AI

The Australian government wants to impose security obligations and enhanced public oversight on the most advanced AI models. The executive aims to establish national safeguards, build trust, and attract investment before any international discussions.

The government prioritizes domestic safeguards before negotiations

Anthony Albanese, Richard Marles, Tony Burke, and Tim Ayres indicate that security safeguards will be implemented at the national level before the opening of international negotiations. The government emphasizes the protection of citizens and national interests. It also aims for the trust and regulatory stability deemed necessary to attract investments. The executive claims it wants to enable AI to bring benefits to the economy and public health. Further details will be provided once consultations are completed.

Independent audits, continuous monitoring, and immediate reporting required

Developers of cutting-edge models will need to open their systems to independent audits. They will be required to maintain continuous security controls and publish their security frameworks. Any security incident must be reported immediately to the government, a requirement designed in part to spot early warning signals of AI agents going off track. A new oversight body, linked to the Australian AI Safety Institute, will enforce these obligations and may impose severe penalties in case of violations.

A framework focused on high-risk models aligned with the national agenda

This system targets the most advanced AI models, identified as having the highest levels of risk. For the first time in Australia, the government has introduced a bill that establishes security obligations for leading AI companies operating in the country. This initiative is part of the agenda "AI in the Interest of Australia."