The decision to block foreign nationals from accessing Anthropic’s most advanced artificial intelligence models, Mythos and Fable, has triggered outrage in tech circles. It signals that AI regulation is no longer about safety or ethics alone; it is now entangled in geopolitics – and the system governing it is looking seriously dysfunctional.
The US adminstration now views frontier AI as a strategic asset akin to Nvidia’s AI computer chips, nuclear technology and advanced weapons systems. Under President Donald Trump, access to the most powerful models is a matter of who is “inside” and who is “outside”. If the US considers you a “foreign national”, you’ll miss out.
That shift exposes a growing incoherence in how AI is regulated globally. For the past few years, we’ve been told AI governance should be collaborative, international and values driven. Governments have convened summits, signed accords like the Bletchley Declaration and talked up shared responsibility. Yet when it comes to the crunch, when the most advanced capabilities are on the line, co-operation gives way to control.
The US, supposedly, has a regime where government officials vet models for safety and national security interests before they are released. The system is clearly broken.
China, of course, has long treated AI as a state-directed strategic industry. The US is now playing the same game. There are even reports in US media that Trump may require the large US AI companies to cede equity stakes to the government, a sort of partial nationalisation of the tech.
The result is a bifurcating AI landscape, where access to cutting-edge systems is determined less by merit or partnership and more by nationality and alignment.
For countries like New Zealand, that should ring alarm bells. If access to leading models can be switched on and off at the stroke of a policy decision in Washington, then our dependence on foreign AI providers starts to look like a serious vulnerability.
It’s tempting to dismiss this as a problem for Big Tech firms or research labs. It isn’t. AI will be embedded in everything from public services to business operations to scientific research. If access is restricted, delayed or politically conditioned, the downstream effects will be felt across the economy.
So what’s the alternative? The answer is not to try to outspend Silicon Valley or compete directly at the bleeding edge. That would be futile. Instead, we should focus on building sovereign AI capability grounded in open-source models.
The open-source ecosystem – from Meta’s Llama family to a growing array of community-driven models – has matured rapidly. These systems may not match the absolute frontier performance of closed models like Mythos or Fable, but they are increasingly “good enough” for a wide range of real-world applications – document analysis, customer service, coding assistance and research support.
Crucially, they can be run, adapted and governed locally. It means we can retain control over our data, tailor systems to local needs and avoid being at the mercy of geopolitical shifts. It also opens the door to building local expertise – the engineers, researchers and policymakers who understand how these systems work and how to deploy them responsibly. Kiwis’ trust in AI is low by international standards. Having more autonomy and control will help build confidence that the technology can work for us.
The Anthropic controversy should be a wake-up call. The era of frictionless, global access to the best AI tools may be ending. In its place, we are seeing the emergence of an AI landscape shaped by national interests and strategic competition. NZ cannot afford to be a passive consumer in that world.