Japanese startup Sakana AI launched Fugu Ultra, an orchestration model that routes tasks across a swappable pool of frontier LLMs, shortly after US export controls disrupted foreign access to Anthropic's most advanced Fable and Mythos models. The timing was hardly coincidental. Fugu Ultra exists specifically to help customers switch among AI systems when one becomes unavailable — a product built on the premise that American AI will become less reliable over time.

That premise, argued Clark D. Asay, Terry Crapo Professor of Law at Brigham Young University, in an analysis for The National Interest, reflects the core commercial risk now facing US AI vendors: unpredictable access to American products creates market opportunities that foreign competitors are eager to fill, including increasingly capable Chinese models.

The switching-cost problem

The stakes extend beyond any single product launch. Asay frames AI as the next technological inflection point, comparable to the internet, smartphones, and cloud computing — platforms where American companies became the world's defaults and captured the resulting productivity gains, innovation, and economic dynamism. The countries and companies that establish themselves as foundational AI providers, whose models become critical infrastructure for businesses and governments worldwide, will shape the trajectory of technological innovation for a generation.

Switching dynamics amplify the risk. Companies and governments that structure operations around a given AI platform tend to stick with it. The providers that win users today are likely to keep them for years, and reliability — not just capability — often drives those decisions. Washington, Asay argues, may currently be making American AI less reliable.

RASA widens the aperture

Pending legislation could compound the problem. The Remote Access Security Act (H.R. 2683) in the 119th Congress aims to restrict adversaries' remote access to American computing infrastructure — a goal Asay does not dispute. But as currently written, RASA's scope is broad enough to burden allies and commercial users who pose no meaningful risk, leaving them little choice but to build their own solutions or turn to non-American AI providers.

The strongest counterargument, Asay acknowledges, is that short-term disruption is an acceptable price for keeping transformative technology away from adversaries. His rebuttal: restrictions that are broad, unpredictable, or poorly targeted may do more to accelerate the development of non-American alternatives than to eliminate genuine threats. That outcome shrinks the global footprint of US technology while weakening America's long-term strategic position.

Narrow, predictable, focused

Asay's prescription is specific. Export control policy should be narrow, predictable, and focused on real threats. Treating disruption abroad as an acceptable byproduct — or as someone else's problem — is a mistake the United States cannot afford, he writes, because the foundational decisions about AI adoption are being made right now.

The author brings a practitioner's background to the argument. Before joining BYU, Asay taught at Penn State's Dickinson School of Law and, prior to legal academia, worked at Amazon's Lab126, supporting the Kindle, Kindle Fire, and Amazon Fire teams. His research focuses on the intersection of law, technology, and innovation.

For AI vendors and their supply chain partners, the analysis lands at a moment when the commercial consequences of policy are becoming visible in shipped products rather than white papers. Sakana AI's bet — that customers will pay for insurance against American supply disruption — is now a real offering in the market. If pending legislation such as RASA passes in its current broad form, the competitive dynamics it may set in motion, with allied customers diversifying toward non-American models while they still can, will be difficult to reverse.