EU probes Anthropic model access cut for European users

Brussels is testing how much leverage it really has over a U.S. AI company after Anthropic cut European users off from its most advanced models under a U.S. export-control directive. The European Commission said it was still in contact with Anthropic and was examining the practical consequences for European users, turning a compliance move into a larger question about who in Europe gets access to frontier AI tools.

Commission spokesperson Thomas Regnier said contingency measures should not be discriminatory against partners, signaling concern that Washington’s national-security rules can spill directly into European research, business and public-sector planning. The Commission also said the newest AI models can support cyber-defense while also raising serious cybersecurity concerns, placing the issue at the intersection of safety, trade and digital sovereignty.

Anthropic said on June 12 that it was suspending access to Claude Fable 5 and Claude Mythos 5 to comply with a U.S. government directive that applied to foreign nationals, including its own foreign-national employees, whether they were inside or outside the United States. “We are suspending access to Claude Fable 5 and Claude Mythos 5,” the company said. By the following day, users were unable to access the models.

The disruption matters because Fable 5 had been described in Anthropic’s June 9 launch materials as state-of-the-art on nearly all tested benchmarks. Mythos 5 was positioned as a higher-trust system for a small group of cyberdefenders and infrastructure providers through Project Glasswing, a program run in collaboration with the U.S. government. Anthropic priced both models at $10 per million input tokens and $50 per million output tokens, underscoring their commercial significance before the cutoff.

The episode has also highlighted how far U.S. controls have moved beyond earlier restrictions centered on semiconductor chips. Previous export controls had focused on the hardware powering AI systems, not the models themselves. Anthropic said it had received only verbal notice of a potential narrow jailbreak and disputed that this should have triggered a recall, leaving Europe with a sharp example of how quickly geopolitical controls can reshape access to advanced AI.