Australia sets mandatory safety rules for frontier AI companies

The government of Australia told about a law for mandatory safety rules for frontier AI companies that work in the country. Prime Minister Anthony Albanese leads the government. This is the first time. The rules target the models that have the most risk. A new body will make companies obey the rules and give large penalties.

What the law requires

The new body will work with the Australian AI Safety Institute. Companies that make frontier AI models must:

• open the models to audits by external parties
• keep up safety checks
• publish their safety frameworks
• tell the government about incidents immediately

The proof

The announcement gives no figure for the penalties and no test results. The incident rule is also for the first signs of AI agents that act in a wrong way.

What is not known

The government gives more information when the consultations end. The announcement does not tell which models have the most risk. The ministers tell that the government makes rules for the country before talks with other countries.

Insight: Australia sets its rules before the world talks

Australia wants audits by external parties and incident reports immediately. A New York bill of October 2026 asks for an audit by a licensed examiner each year and a report in 72 hours. Australia goes first, before talks with other countries, which reads as a move to set the rules that other countries copy. For leaders: the same duties are possible in more countries.