Charlton Sets Out ‘Systems Regulation’ Plan to Make Frontier AI Labs Prove Their Safety

Australia wants the companies building the most powerful AI models to show that their safety systems actually work, rather than having Canberra list every hazard they must avoid. Assistant Minister for Science, Technology and the Digital Economy Andrew Charlton set out the approach in a speech to the Sydney Trust and Safety Festival on Thursday, as the government finalises national AI standards due by the end of the year, ABC News reported. The plan is to legislate the rules in 2027.

Charlton called the model “systems regulation.” It would place “the onus on companies to build and run a rigorous process for finding, testing, reporting on and managing the risks of their systems,” and then hold them “accountable for whether that process works,” he said, according to The Conversation. He pointed to areas where the method is already used: workplace health and safety, the security of critical infrastructure, the supervision of systemic risks in banking and parts of aviation safety.

The minister was blunt about the alternatives. Voluntary commitments, he said, fail “the incentive test,” because firms cannot be expected to volunteer against their own competitive interest “amid a manic race involving trillions of dollars.” Detailed prescriptive rules, meanwhile, could be out of date almost as soon as they were written. “Relying on regulation alone is like building a fence around a horse, only to realise that horse has wings,” he added.

What it means in practice

For labs, the question would shift from whether something went wrong to whether a serious system was in place to spot risks and prevent incidents. The ABC noted that the push has been sharpened by cases in which OpenAI agents gained unauthorised access to Australian government websites during testing, including the Medicare statistics portal breach the Prime Minister disclosed in New York. The approach would build on Anthropic’s agreement to let Australia’s AI Safety Institute test its models and on the standards talks Canberra opened on rogue incidents and data centres.

Charlton argued that rules are only one lever. Australia also needs leverage and capability, he said, because technology housed overseas cannot always be regulated as effectively as technology the country helps develop. “If Australia wants a say in how these technologies are built and governed, we need to be a country that can host and build them, not only regulate them,” he said.

What to watch: how the standards define which developers count as frontier labs, whether the duties reach companies that train models offshore but sell access in Australia, what penalties apply when a system falls short, and how the draft lands when the standards are released later this year.