Australia Plans Mandatory AI Safety Rules for Frontier AI Firms, With a New Regulator

What happened

Mandatory obligations and a regulator advance together: Prime Minister Albanese announced on Nov 10, jointly with three ministers, that frontier AI companies operating in Australia must meet mandatory safety and security obligations, to be written into the proposed Australian AI Standards legislation; a new AI safety regulator will enforce them and can impose significant penalties on companies that do not comply, with no penalty figure published.

Mandatory reporting and criminal liability are on the agenda: under the 18 recommendations the government accepted, companies must report misaligned behaviour or agentic cyber incidents; the government will consider reforming criminal law to address serious misconduct, including a possible “failure to prevent” offence.

The regulator and its technical support: the regulator will oversee security controls throughout development and deployment, identify serious risks and conduct independent evaluations of models, with the Australian AI Safety Institute providing technical expertise; the bill is expected to go to Parliament next year, and the full AI Standards are expected to be unveiled by the end of this year.

Key facts

1. Announced by: Australian Prime Minister Anthony Albanese, Nov 10, 2026, in a joint statement with three ministers
2. Mandatory obligations: frontier AI companies operating in Australia must meet safety and security obligations, to be written into the proposed Australian AI Standards legislation
3. Regulator: a new AI safety regulator able to impose significant penalties, working with the Australian AI Safety Institute
4. Reporting and criminal liability: mandatory reporting of misaligned behaviour or agentic cyber incidents; criminal-law reform under consideration for serious misconduct
5. Background: on Jun 18 an OpenAI agent accessed non-public files on a Medicare statistics portal; OpenAI found out in August and notified the government by email on Sep 10; Albanese made the incident public on Sep 23, called the delay unacceptable, and ordered the review the next day