Trump’s aborted AI order shows who writes the AI rules

Australian chief information officers and policy teams buying frontier AI models do not need to wait for Canberra’s next consultation to see where the market is moving. After Reuters reported that Donald Trump postponed a planned AI executive order, the signal from Washington was blunt: the biggest US labs still have room to push back when a safety regime starts to look like a gate on product release.

That matters in Australia because the models sold here are largely designed, trained and defended offshore. The White House had considered a voluntary framework that would give the government access to frontier systems up to 90 days before release, according to an Axios draft of the order. A second Axios report that published the seven-page draft showed how close the administration had come to formalising that approach before shelving it. For Australian buyers, the practical question is less whether Washington will write a hard federal rulebook than what evidence local customers will demand if the US decides the lightest touch is the safest commercial one.

“I didn’t like certain aspects of it, I postponed it. We’re leading China, we’re leaving everybody, and I don’t want to do anything that’s going to get in the way of that lead.”

• Donald Trump, Reuters

Across the frontier-model lobby, the same draft was read differently. The Guardian’s analysis of the reversal framed the collapse of the order as a victory for tech influence, while Politico’s look at OpenAI’s state strategy suggested the lobbying centre of gravity had already moved beyond the White House. If Congress stalls, industry leans on the states; if states move, companies try to shape the language. The tension between buyers who want assurance and vendors who want speed sits at the centre of Trump’s reversal.

A federal retreat still leaves a power fight

This was not a European-style AI Act in waiting. On the reporting so far, the cancelled order was a compromise document: voluntary in form, but still significant enough to worry companies that fear any pre-release access regime becoming a template for later scrutiny. That is why the 90-day window mattered. A rule sold as light-touch can still set expectations about testing, documentation and who gets to ask hard questions before launch.

Reuters quoted Elon Musk denying he had pressed Trump to drop the order, even as Axios reported that more than 60 MAGA allies had urged the White House to require some form of vetting before advanced AI systems reached the public. That split matters. Some figures in Trump’s orbit wanted fewer barriers, but others wanted controls aimed at release-time testing rather than ongoing federal supervision. The argument was never simply regulation versus no regulation. It was about who gets to write the least painful version.

The battle is also moving into adjacent arenas. Wired reported that former OpenAI staffers are warning investors about xAI’s safety record, while The Verge argued that Anthropic and OpenAI have started taking their rivalry into electoral politics. A softer White House stance does not end the governance fight. It pushes more of it into courtrooms, campaign networks and private negotiations, where well-capitalised labs are usually strongest.

The states are now the real battleground

Statehouses are the next venue. Politico’s reporting on OpenAI’s statehouse campaign showed Chris Lehane, the company’s chief global affairs officer, working blue-state capitals partly because they still have the power to write the first enforceable rules that affect frontier models.

“It was pretty clear that those states … I don’t think this was a question of whether regulations were going to happen or not. It was what kind.”

• Chris Lehane, Politico

Seen through that lens, the aborted order looks less like a failed one-off and more like a fight over templates. The next phase is unlikely to be a clean federal settlement. It will be a contest over templates, state by state, procurement rule by procurement rule, lawsuit by lawsuit. Genevieve Smith argued in the Guardian that the fight over Colorado’s anti-discrimination law showed how quickly federal rhetoric can be deployed against state experimentation. Trump’s reversal makes that pressure more potent because companies can now claim the White House itself saw heavy-handed oversight as a competitive drag.

Australia is unlikely to copy the US retreat

Australia’s starting point is different. APRA’s letter to industry on AI already treats AI as a governance, risk and accountability issue for regulated entities, not as a future political abstraction. The Victorian Law Reform Commission’s overview of AI regulation notes that Australia still lacks AI-specific legislation, which leaves sectoral guidance doing most of the real work. That is messier than a single AI statute, but it is not the same thing as a hands-off posture.

Here, the burden on enterprise buyers and policy teams is more likely to move sideways than disappear. If US labs win more freedom at home, local customers will still want model documentation, testing records, internal accountability lines and clearer statements about where a vendor’s safety claims stop and marketing begins. In other words, the artefacts that the cancelled US order might have normalised through Washington could instead be demanded through procurement, prudential supervision and board risk committees.

For Canberra, the choice is narrower but more concrete than the US debate suggests. It can keep waiting for a comprehensive AI law that may take years, or it can keep hardening sector-specific expectations around the systems already entering banks, government agencies and large enterprises. Trump’s aborted order makes one point clear for Australian policymakers: if the US federal government is unwilling to hold the line, buyers and regulators further down the chain will have to decide how much proof they want before they trust the next model release.