The Speed of Trees

In The Lord of the Rings, two hobbits climb into the branches of Treebeard, an ancient walking tree, and try to convince him that his forest is in danger. An army is cutting it down at the edges. Treebeard is wise, and he is not unwilling. He is simply slow. It takes him most of a day to say hello. By the time the trees have finished deliberating, the question is whether there will be a forest left to save.

Dario Amodei, the CEO of Anthropic, opened his latest essay with that image. Policy on the AI Exponential, published this month, is his most direct argument yet that governments move at the speed of trees while AI moves at the speed of light, and that the gap between the two is now the central governance problem of our time.

We read it closely, because the question underneath it is the question our organisation was built on. If the rules cannot keep pace with the technology, what holds trust together in the meantime?

What the essay actually says

Amodei's argument runs across five fronts: public safety regulation, the economics of job displacement, accelerating AI's benefits in fields like medicine, protecting civil liberties, and geopolitics. But the heart of it is a shift in his own position. For years, Anthropic advocated transparency as the right first step, on the basis that legislating too early risks regulating the wrong things. That era, he argues, is over. The risks are no longer hypothetical. Recent demonstrations of frontier models' cyber capabilities have shown that advanced AI systems are now matters of genuine strategic consequence, and he believes biological and autonomy risks may follow.

His proposed answer is modelled on aviation. Just as aircraft must pass independent testing before they carry passengers, frontier AI models should pass independent testing before they reach the public, with government holding the power to block deployment that fails the standard.

Then comes a detail that most coverage of the essay skipped past, and the one we think matters most for Australia. Amodei suggests the testing itself need not be done by a government agency at all. It could be done by private organisations, authorised and inspected by government, assessing against published standards. Policy researchers call this a regulatory market. Government sets the bar. Independent bodies hold people to it. The system moves at the speed of business while remaining accountable to the public.

Keep that idea in mind. We will come back to it.

Where Australia actually stands

Here is the part of the story that surprises people. While the United States and Europe move toward binding rules, Australia has deliberately chosen a different path.

In 2024, the federal government proposed mandatory guardrails for high-risk AI. The tenth of those proposed guardrails required organisations to undergo conformity assessments, in other words, independent certification that the rules were being followed. Then, in December 2025, the National AI Plan set that proposal aside. Australia would rely on existing laws, sector regulators and voluntary guidance instead. The National AI Centre's Guidance for AI Adoption, with its six essential practices, became the government's primary reference point, and a new Australian AI Safety Institute was funded to test systems and identify where the genuine gaps are.

Reasonable people disagree about whether that was the right call. The Productivity Commission warned that AI-specific legislation could chill innovation. Critics counter that gap analysis takes years while the technology compounds monthly. Both can be true at once.

But notice what the choice creates. Australia now has government guidance without government enforcement. Principles without proof. Every Australian business is being asked to do the right thing with AI, and almost none of them have a way to show anyone that they have.

The gap is not regulatory. It is trust.

While the policy debate continues, something else has shifted that no consultation paper fully captures. The public has made up its mind that AI risk is real. Amodei himself rejects the idea that this is a marketing problem for the industry to spin away. People are worried, he writes, because they correctly perceive the risks, and that concern is democratic accountability working as intended.

For Australian businesses, this lands as a practical, commercial fact. Your customers are now asking questions they were not asking eighteen months ago. Who reviews your AI before it touches my data? What happens when it gets something wrong? Can anyone outside your own marketing department vouch for you?

In a market with binding regulation, the answer is compliance. In a market without it, the answer has to be demonstration. And demonstration only counts when it comes from someone other than yourself. A business marking its own homework is not assurance. It is advertising.

The infrastructure has to exist before the law arrives

This brings us back to the regulatory market idea, because there is a quiet assumption inside it that deserves to be said out loud. A regulatory market only works if the independent assessment bodies already exist, with standards, processes, registers and track records, on the day government decides to rely on them. You cannot legislate an assurance ecosystem into existence overnight. It has to be grown in advance, the way airworthiness inspection grew up alongside aviation long before it was universally mandated.

That is the work we believe Australia should be doing now, and it is the work we have set out to do. Our certification framework, lodged with IP Australia and the ACCC as a registered certification trade mark scheme, assesses Australian businesses against the Australian AI Ethics Principles across three tiers of maturity: a public commitment, an operational policy, and a full governance framework with risk management, oversight and accountability built in. Every certified business is independently assessed, publicly listed, and reassessed annually. The rules are filed, published and enforceable.

We are not waiting for legislation to make responsible AI verifiable. We are building the proof layer that voluntary guidance needs to mean something, and that future regulation, whenever it comes and whatever shape it takes, will need to stand on.

Amodei describes AI as potentially becoming a country of geniuses in a datacenter. Whether or not that arrives on his timeline, Australian businesses are already deploying systems they did not build, do not fully understand, and cannot personally vouch for. The honest response is not paralysis, and it is not blind adoption. It is governance you can show.

What Australian businesses can do this quarter

You do not need to wait for Canberra, and you should not. Map your AI use honestly, including the tools your teams adopted without asking. Put a written policy behind it, one your staff actually follow rather than a template no one reads. Assign a named owner for AI oversight, because accountability that belongs to everyone belongs to no one. Align what you do with the National AI Centre's six essential practices. And then take the step that turns all of it from a claim into a credential: have someone independent verify it.

That is what certification is for. Not a plaque on a wall, but a standing answer to the question your customers are already asking.

The forest is waking

Amodei ends his essay on a hopeful note. Treebeard, he observes, is finally stirring. Policymakers around the world are more open to serious action on AI than they have ever been, and the window for getting it right is open now.

He is right, and we share the optimism. But hobbits did not wait in the branches for the trees to finish deliberating. They made the case, they moved, and the forest followed.

Australian businesses face the same choice. You can wait for the slow, careful machinery of government to tell you what responsible AI requires. Or you can demonstrate it now, verifiably, while trust is still there to be won.