Australia's planned AI safety laws: what's coming by 2027?

Canberra wants the most powerful AI built on proven safety systems, borrowing ideas from banks and airlines, so Australians can use it with confidence.

Australia AI safety laws brief

0:00

/100.403084

The short version

On 8 October, Assistant Minister Andrew Charlton gave a speech in Sydney about how Australia plans to regulate the most powerful AI. According to the ABC, the federal government is finalising national AI standards due by the end of this year, with plans to put the new rules into law in 2027.

The big idea is simple. Instead of trying to write a rule for everything an AI might do, the government wants AI developers to show they have a serious system for finding and fixing problems, and that the system actually works. The goal is trust, so more people and businesses feel confident using AI.

Wait, which AI are we talking about?

This is aimed at what the government calls frontier AI: the most advanced models, built by the biggest AI labs, with the most serious potential harms. Charlton's speech was about "frontier AI regulation when it comes to the most significant harms".

So no, this is not a new rulebook for the local bakery's chatbot. If you run a small business, nothing changes for you tomorrow. But the thinking behind it is worth understanding, because it will shape how AI tools are built and sold in Australia, and it should make choosing them easier.

Borrowing from banks and planes

Charlton said Australia can learn from "systems-based" regulation. In his words, it "places 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 holding them accountable for whether that process works."

He pointed to places where this already happens: workplace health and safety, security of critical infrastructure, prudential supervision of banks, and parts of aviation safety. The government sets the standard. Companies prove they meet it.

His argument for not writing lots of detailed rules was refreshingly practical. Technology moves in months, not years, so detailed rules "could be out of date before the ink is dry."

Why now?

Charlton's argument is about incentives and trust. He said voluntary codes are fast and flexible, but struggle when companies are racing each other: "You cannot ask firms to volunteer against their own competitive interest amid a manic race involving trillions of dollars and then be surprised when they do not."

He also made the positive case. Under this approach, "the cheapest path to market is ultimately the one that runs through safety, not around it. Because when people trust your model is safe to use – they're more likely to use it."

According to the ABC, recent incidents involving AI agents in testing helped sharpen the push, and OpenAI has since backed mandatory safety requirements, independent assessments and incident reporting. In other words, at least part of the industry and the government are heading in the same direction.

The best line of the speech

Charlton was also refreshingly honest that rules are only one part of the plan. His metaphor: "Relying on regulation alone is like building a fence around a horse, only to realise that horse has wings."

His answer is that Australia also needs leverage and capability, meaning the ability to host and build AI here, not just regulate it. That is an exciting ambition for Australian AI.

What did the other side say?

According to the ABC, Shadow Defence Minister James Paterson said the position Charlton outlined was heading "in the right direction", adding that "the details will matter" and the Coalition would not give Labor a "blank cheque" on AI.

What it means for you

If you are a regular AI user or a small business owner, here is what it means in practice:

Nothing changes overnight. Standards are due by the end of 2026, with legislation planned for 2027.
The target is the biggest AI developers, not everyday businesses using AI tools.
Expect more evidence about testing. Over time, the AI products you buy should come with clearer information about how they were tested, which makes choosing tools easier.

And there is a habit worth borrowing now. The question Charlton asked of AI developers works for anyone using AI: is there "a serious system in place to detect risks and prevent incidents occurring"? For a small team, that can be as simple as a short checklist and a named person who owns it, which is exactly what lets you use AI for more.

Where Walter fits

Walter is model-agnostic, and we welcome clearer expectations that help everyone use AI with confidence. Walter is being built to pull your business context together across the AI tools you choose, with clear owners and records, so your team can make the most of AI as the rules take shape.