US order blocks foreign access to Anthropic

A critical AI system can vanish overnight while the servers are running and the invoices are paid. On 12 June 2026, Anthropic said the US government ordered it to suspend foreign-national access to Fable 5 and Mythos 5, including for its own foreign-national employees. For Europe, the shock is practical: as AI becomes operating infrastructure for banks, telecoms, labs and administrations, a foreign export-control switch can become a domestic outage.

The Safety Claim Came First

The order was reportedly triggered by a claimed bypass of Fable 5’s safety barriers. A bypass, or “jailbreak”, is a way of wording requests so the model steps around instructions it is supposed to obey. Think of a secure building where the wiring works, but someone talks the front desk into handing over the wrong badge.

That matters because these models were designed as guarded rooms. Business Insider reported that Anthropic built Fable 5 to restrict requests in cybersecurity, biology and chemistry. Mythos sat closer to the restricted side: Le Monde had already placed it inside French cybersecurity discussions involving Orange and Crédit Agricole.

The public case is thinner than the consequence. In its own review, Anthropic said it understood the government had seen a claimed bypass, but judged the demonstration to involve minor, previously known vulnerabilities. AP reported that the Commerce Department gave no public restoration timetable or detailed explanation. National-security officials may have reasons they cannot publish. Businesses saw the visible result first: access disappeared.

Nationality Became The Circuit Breaker

The legal logic is older than AI. Under US export rules, sharing controlled technology with a foreign national can count as an export even if the person is sitting in California and the data never leaves the cloud, as federal rules set out. Applied to a live AI service, that turns identity into infrastructure.

For a cloud model, nationality is hard to separate cleanly. Access runs through customer accounts, employees, contractors, API keys, regions and support systems. R Street argued that without a reliable way to check citizenship at every doorway, the simplest compliant move is to close access until the rule is clear. Reuters-linked reporting said Anthropic disabled access after the order and that AWS was asked to revoke access across regions.

For a bank, ministry or telecoms operator, the resilience question is blunt: if one legal order removes your main model tonight, which process still works tomorrow morning?

France understood the point because it had real users in view. Le Monde also tracked an unusually broad political reaction, with French figures from Jordan Bardella to Jean-Luc Mélenchon, Gabriel Attal and Édouard Philippe reading the move as proof that AI dependence had become a strategic exposure. Brussels used calmer language, but Euronews reported Commission spokesperson Thomas Regnier saying the EU was assessing consequences for European users.

Europe Needs Backups, Not Slogans

The fastest European answer is architectural: avoid single-model dependence, keep fallback providers, demand that data and prompts can move, and decide which workflows are too sensitive to rely entirely on foreign-controlled systems. Germany’s debate is already there. Tagesschau covered the incident as a resilience test for administrations and critical businesses, with US AI still in the procurement picture.

EU policy now lands in a sharper context. The Commission presented its tech-sovereignty package on 3 June 2026, just days before the Anthropic shock. The Cloud and AI Development Act proposal says EU data-centre capacity should at least triple within five to seven years. InvestAI adds the industrial bet, with €200 billion for AI development and €20 billion for AI gigafactories.

Those plans can build leverage over time. They do not change who controls a proprietary US model today. Europe can regulate AI, buy computing capacity and fund local champions, but it also needs the unglamorous habits of critical systems: exit routes, backups, contract clauses, open-source options where possible, and public services that keep running when one foreign switch flips.