The Trump administration and AI giant Anthropic are at it again. In the latest stoush, the administration claims that security risks in the company’s new models have forced Washington to apply export controls, banning their use by foreigners.

This should worry Australia because it points to a trend that has been building for months. As advanced AI models become more powerful and more critical to national security, they are being treated as guarded national assets, not to be shared even with trusted allies.

Australia must work urgently to shore up access to cutting-edge AI that will, like no technology before it, transform our economy, society and security and determine our place in the world. The good news is we have cards to play.

AI commentators are buzzing with theories about the Trump team’s true motivations. The administration claims the new models can be jailbroken, meaning malicious actors can use clever prompts to get around the models’ safeguards and trick them into doing harm. This would be a security threat. But Anthropic protests that any jailbreaks found are minor and typical of frontier models. They have a point: Britain’s revered AI Security Institute has found such narrow jailbreaks in every previous model.

If these new models are the step-changes they seem to be, it makes sense for the administration to hold tightly onto them. After all, if your companies are building superweapons, why would you share them with other countries? That’s true even if the administration wants to rein in a company it hates, as some commentators claim. The policy might seem haphazard and fumbling, but it’s still headed in the same direction.

Evidence for this trend is visible in the fact that the National Security Agency – the United States’ cyberspies – have reportedly been using Anthropic’s hacking tool Mythos to carry out offensive cyber operations against foreign adversaries. Australian entities, including our highly respected Australian Signals Directorate, managed to negotiate access to the formidable and restricted Mythos model to check for cyber vulnerabilities. We need to make sure such access continues even as models grow more powerful.

To give ourselves the best chance, Australia should take three steps. First, we should carve out a commanding place in AI development – most obviously by building data centres to train models – so that the Americans need us as much as we need them.

Companies such as Anthropic are clamouring to build data centres here, thanks to our abundant land, renewable energy, political and regulatory stability, and distance from threats such as Iranian missiles and drones. This puts Australia in a strong position and gives us leverage over frontier models in the future.

Such data centres would also create innovation clusters, for instance by bringing cutting-edge researchers from Anthropic, OpenAI, Google DeepMind and others to work in Australia. Presently, these labs’ offices are largely for sales and government relations.

To clear the way, Australia will need to find a compromise on our copyright laws, which are deterring AI companies from training on Australian territory because the vast quantities of data needed puts them at risk of getting sued.

Second, we should sell Washington on our role in supporting US AI dominance, emphasising our alliance, our Five Eyes partnership, and our ability to help them by training models here and relieving their creaking power grid – a winner with the America First agenda.

Defence industry cooperation is a good paradigm. As the defence-tech start-up Anduril argued on 14 June, the US should loosen export controls so allies can contribute to weapons production at a time when speed and industrial capacity are vital to deterring a rival such as China.

Cooperation under AUKUS on national security technologies, including AI, is an excellent basis, as are examples such as the F-35 Joint Strike Fighter project in which Australia built parts for the global supply chain.

If the latest hit on Anthropic is just politically driven lawfare, as the AI expert and former Trump official Dean Ball speculated on X, it is an own goal at a time when the US and China are in a neck-and-neck race to dominate frontier development and embed their products around the world. It could drive a brain drain as foreign employees of US AI labs are legally excluded from working on the latest innovations.

With such uncertainties about the coherence of US policy, the third measure Australia should take is to diversify our options by deepening AI industry cooperation with like-minded middle powers that have some serious game in the technology. Europe, Japan, South Korea and Canada are all natural partners. Together, we can build solid AI ecosystems in the event that the US becomes mired in techno-nationalism.

Whatever the machinations in Washington, the latest move fits a trend towards greater national control of the most powerful technology the world has ever seen. Australia must do all it can to make sure we are not left behind.

This article was originally published in the Australian Financial Review.