Twelve months ago, the Productivity Commission told the Australian Government not to impose mandatory guardrails on AI. Existing laws would do. Don’t chill innovation.
Two weeks ago, the Prime Minister stood up in Sydney and reversed the lot.
Australia is moving from voluntary instruments – the AI Ethics Principles of 2019, the Voluntary AI Safety Standard of 2024 – to a mandatory, whole-of-government framework, with a dedicated Office of AI and legislation expected early next year.
I shared my first reaction last week: nothing in this framework makes your organization better at using AI, and the real question for boards is whether you’ll be compliance-ready or AI-ready when it lands. The response to that post told me something useful. The question executives are actually sitting with isn’t whether the distinction matters. It’s more practical: what exactly is coming, and what do we do between now and 2027?
So this edition is the brief I’d want in your hands before the inevitable “Australian AI Standards – Implications” paper arrives in your board pack. The ninety-second version of what was announced, the sleeper issue almost nobody is covering properly, and the five questions that will make you the most useful person in the room when the compliance conversation starts.
The policy, in ninety seconds
What happened: On 15 July, the Prime Minister delivered a speech titled “AI in Australia’s interests” announcing two things: a new Office of AI inside the Department of the Prime Minister and Cabinet, operational immediately, and a set of mandatory Australian AI Standards.
The world-first claim: The Government says the Standards will make Australia the first country to bring the economic, social, national security and environmental dimensions of AI into a single national framework.
The timeline: National Cabinet considers the Standards next month. Legislation goes to Parliament early next year. So: mandatory obligations, most likely in force during 2027.
Data centres carry the heaviest obligations: operators will be required to underwrite their own new power supply (including renewables) without passing costs to consumers, be net generators of energy, pay for their own grid connections and water infrastructure, reduce water use and lift energy efficiency – with fast-tracked approvals as the trade-off.
The reversal is the story: this is a government that was advised, in writing, by its own Productivity Commission, to stay hands-off – and chose the opposite. Whatever else you take from the announcement, take this: the era of voluntary AI governance in Australia is over.
That’s the frame. Now the part your board pack probably won’t cover properly.
The sleeper issue: copyright just became a business risk conversation
Most coverage treated the copyright announcement as an arts-sector story. I think that’s a serious misread.
Here’s what was actually said. The Government has again ruled out a text and data mining exemption – the legal carve-out AI companies wanted so they could train models on Australian works without permission. The PM went further: Australian works cannot be used to train AI systems without the artist’s control, including control over price. Anything less, in his words, amounts to theft.
Now here’s why that’s a boardroom issue and not just a creative-industries one.
First, the mechanism is completely undefined. The Attorney-General is still consulting on how creators actually get paid, by whom, and for what. That uncertainty cuts both ways: if your organization creates content – reports, research, media, training material – the new settings may become an asset. If your AI use depends on models trained on data of uncertain provenance, the “amounts to theft” framing should sharpen your due diligence questions now, not in 2027.
Second, the stakes are enormous and very current. Anthropic’s proposed $21.6 billion investment in Australian data infrastructure has been publicly reported as contingent on how this copyright question resolves. When a single AI company’s Australian commitment – one of the largest tech investments this country has ever been offered – hinges on the same policy your business will be governed by, “wait and see” stops being a neutral position.
Third, and least discussed: every organization is now on both sides of this. You consume AI trained on other people’s work. You produce work that AI companies would happily train on. The board that only asks “are we allowed to use these tools?” is asking half the question. The other half is “what is our own content worth in this new settlement, and who’s protecting it?”
Nobody’s briefing boards on that second half yet. Be the person who raises it.
Five questions before the compliance papers arrive
Between now and 2027, legal and risk teams across the country will – quite properly – swing into action. Compliance workstreams will spin up. AI will get plenty of boardroom airtime, and almost all of it will go to one question: are we compliant?
These five make sure the more important conversation happens alongside it:
- Who owns AI capability in this organization – as distinct from AI compliance? If the same person owns both, capability will lose. It always does.
- What would the Standards have to say for us to change our AI strategy? If the honest answer is “very little,” then waiting for them is a performance.
- Do we know our copyright exposure – in both directions? What trained the models we use, and what is our own content worth under the new settings? (See above – almost nobody is asking the second half.)
- Are our managers equipped to lead adoption, or just to enforce policy? The evidence is blunt: adoption follows the direct manager. A framework can mandate guardrails. It cannot mandate enthusiasm.
- What’s our capability position on the day the legislation passes? Work backwards from that date. Eighteen months is enough time to build genuine workforce AI capability – or to produce a very thorough set of compliance documents. Pick deliberately.
The bottom line
I think the Government has read the moment correctly: Australians want AI managed, and a single coherent framework beats the regulatory patchwork everyone else is wrestling with.
But a national framework sets the floor. It will never set your ceiling.
The organizations that win the next decade won’t be the ones that complied most thoroughly with the Australian AI Standards. They’ll be the ones that treated the same eighteen months as a capability-building window – and turned up in 2027 both compliant and fluent.
The framework is Canberra’s job. The ceiling is yours.
Boards and C-suite: if the Standards have put AI on your agenda and you want the capability conversation alongside the compliance one, I run executive sessions on exactly this.
Conference and event planners: if your 2026-27 program needs an AI keynote that goes beyond the policy headlines – including the live demonstration that shifts a room from fear to strategy.
