The United States government has ordered AI company Anthropic to block all non-American access to two of its advanced AI models — Fable 5 and Mythos 5 — citing national security concerns, leaving organisations worldwide scrambling and putting questions of AI sovereignty at the centre of the G7 summit in Evian, France.

The models, designed to identify security vulnerabilities in software systems, were released last week before being taken offline over the weekend after Anthropic received a letter from the US Department of Commerce. The letter, made public by Bloomberg, stated there was an "unacceptable risk" that adversaries including Russia and China could exploit the technology. Rather than restricting access selectively, Anthropic — unable to technically comply in any other way — disabled the tools entirely, meaning even American users lost access. The company has argued that its built-in safeguards prevent misuse, but Washington remains unconvinced, reportedly after an internal Amazon study — not made public — suggested those protections could be circumvented. Anthropic staff flew to Washington over the weekend for talks, which continued into Monday without resolution. President Trump said negotiations with the company were going "fine."

The abrupt shutdown exposed how deeply organisations in Canada, Europe, Hong Kong, and elsewhere had embedded these tools in critical workflows — and how quickly a commercial relationship can become a strategic dependency. Businesses and institutions that relied on the models found access simply gone, with no appeal process, migration window, or advance warning. EU technology commissioner Henna Virkkunen used the episode to warn that Europe must reduce its technological dependence on other nations; a concern that resonated given that the best AI-powered cybersecurity tools currently come exclusively from the United States or China, with no European alternatives at a comparable level.

For countries like Canada, which hosts roughly 10 per cent of the world's top-tier AI researchers yet deploys less than two per cent of global AI venture capital, the shutdown is a sharp reminder of the gap between scientific leadership and commercial control. Foundational work by researchers such as Geoffrey Hinton and Yoshua Bengio helped shape modern AI, but the commercial benefits — and the power to set terms of access — have largely accrued elsewhere. Most organisations in Canada and beyond license access to AI from a small number of American firms operating under US law, making them subject to US government decisions with no recourse.

The Anthropic case has forced a broader question onto the international agenda: when a technology shapes economic competitiveness, national security, and public services across borders, should access be determined unilaterally by a single government and a handful of private companies? The US controls an estimated 74 per cent of global AI supercomputer capacity, and the frontier models others depend on are largely built and operated by American firms. As the G7 wraps up, diplomats are discussing proposals that would restore allied-nation access to the blocked tools, but no agreement has been reached. Whatever the outcome, the episode has underscored that access to advanced AI is no longer simply a technology procurement question — it is a matter of national and economic sovereignty.