Macron criticises US AI ban as allies seek research access

An AI model ban ordered by President Donald Trump has forced Anthropic to suspend foreign access to two advanced systems, Mythos 5 and Claude Fable 5.

The directive, framed as a national security measure, has quickly strained ties with US allies and raised new questions about cross-border AI research and testing.

US order prompts shutdown of international access

The US government issued the order last week and applied it broadly to foreign nationals both inside and outside the United States. Anthropic responded by taking the two models fully offline to ensure it could comply.

The move is unusual in scope because it effectively blocks global access rather than targeting a specific country, sector, or set of entities. The order arrived as Anthropic was preparing public releases for early June, according to the information provided.

The affected models were described as Mythos 5 and Fable 5, which had been expected to become available publicly in the coming weeks. Anthropic had also been running a related testing program involving its “frontier” model preview.

Security rationale unclear as ‘jailbreak’ concerns emerge

Anthropic said the US government did not provide a detailed explanation for the restriction. The company said it understood the administration believed it had learned of a technique to “jailbreak” Fable 5, a term used for bypassing built-in safeguards.

Before the order, Anthropic had granted access to 200 institutions across 15 countries through a program tied to Claude Mythos Preview. That access was intended to help identify vulnerabilities and stress-test the system’s protections.

The timing is significant for AI safety work, where external testing and “red-teaming” often rely on participation from researchers outside the developer’s home country. Cutting off these channels can reduce feedback loops just as models near public release.

Europe reacts as reliance on US AI comes into focus

The decision reverberated across Europe, where governments and companies lean heavily on US-developed AI tools and infrastructure. The immediate concern is not only access to specific models, but also the precedent of rapid restrictions with limited public explanation.

French President Emmanuel Macron addressed the issue at a Group of Seven meeting this week, describing the US action as both a warning and a setback. He called it a “wake-up call” about AI risks while arguing that imposing limits in this way was “a bad thing.”

Macron also criticised what he characterised as a narrowly national approach. Speaking on Wednesday, he said the response was, in some respects, “strictly nationalist,” underscoring the political sensitivity around who can use frontier AI and under what conditions.

The episode highlights a growing tension between national security controls and the collaborative norms that have shaped international science and technology programs. It also signals that restrictions on AI systems may arrive abruptly and affect allies as well as rivals.

Next steps will hinge on whether the US clarifies the scope and duration of the ban, and whether Anthropic can restore limited access under new safeguards. Governments and research institutions are likely to press for clearer rules that balance safety testing with predictable access.