Anthropic Shutdown Exposes EU AI Dependency Risks

The US Commerce Department issued an export-control directive that, according to a company statement, ordered suspension of access for foreign nationals to Anthropic's most advanced models, Fable 5 and Mythos 5 (Anthropic statement, June 12). Anthropic wrote that because it could not reliably distinguish foreign nationals in real time, it disabled access to those models for all customers worldwide to ensure compliance. The company said the directive arrived at 5:21pm ET and that the government letter did not provide specific national-security details of its concern. AP and The Next Web reported that the shutdown followed claims an unnamed party had demonstrated a jailbreak technique; Anthropic described the demonstration as identifying a small number of previously known, minor vulnerabilities.

Anthropic described Fable 5 as the public-facing variant with safety classifiers and Mythos 5 as the same underlying model with cyber-related safeguards relaxed for vetted defenders, according to The Next Web and Anthropic's own product documentation. Anthropic said it had red-teamed Fable with government and third-party partners for thousands of hours, and that the identified vulnerabilities appear relatively simple and observable in other models without requiring a bypass.

Governments using export controls or emergency directives against commercial model providers create a new form of operational concentration risk for customers who rely on frontier models from a single foreign vendor. Industry observers note that relying on a small set of US-developed models concentrates both technical dependency and policy exposure, which can interrupt services that integrate those models for automation, cybersecurity, or critical workflows, per Politico, Sifted, and Breaking Defense. European political and regulatory actors reacted quickly: Politico reported European Commission spokesperson Thomas Regnier said the case "further underlines Europe's need for technological sovereignty," and several European politicians urged acceleration of local model development. Sifted framed the episode as the first real-world example of a digital "kill switch" with the potential to disrupt services across Europe. Financial Times commentary argued the move could advantage non-US competitors in some markets, including China, while raising questions about access for defence and critical-infrastructure users.

Monitor supplier risk registers, procurement clauses, and continuity plans tied to frontier models. This episode is a real-world case study in how geopolitical or national-security measures can produce immediate service disruption when access to a single vendor's model is treated as critical infrastructure.

• •Any clarifying guidance or appeal from the US Commerce Department about the scope and technical basis for the directive.
• •Whether Anthropic or other frontier-model vendors pursue technical mitigations or policy challenges that would change access controls.
• •Announcements of EU funding, procurement, or public-private efforts aimed at domestic model development.
• •Operational workarounds by enterprise customers, such as model diversification, additional on-premise tooling, or contractual safeguards with providers.