Anthropic shuts Fable 5 and Mythos 5 after U.S. export order
Anthropic says it has taken its Fable 5 and Mythos 5 AI models offline after the U.S. government ordered the company to block access by foreign nationals on national-security grounds. The company says the letter arrived at 5:21 p.m. U.S. Eastern time on 12 June and covered foreign nationals inside and outside the United States, including Anthropic employees. To avoid accidental non-compliance, Anthropic says it disabled the models for all customers while it seeks restoration of access. The U.S. concern, as Anthropic describes it, appears to involve a possible jailbreak of Fable 5, but the company says the demonstrated issue was limited and did not show unique capabilities beyond other public models. The case matters beyond one vendor because it tests whether frontier AI models are now treated less like ordinary software services and more like controlled strategic technology.
Evidenced on the trust ledger·📚 9 sources·🧠 AI-checked·🇧🇪 Belgian: MediumWhy you can trust this
About this story
Anthropic (U.S. AI company founded in 2021 and best known for Claude) is one of the leading frontier-model labs competing with OpenAI, Google DeepMind and Meta. Fable 5 (Anthropic's newly released public model, according to the company) is described by Anthropic as a safeguarded version of the more restricted Mythos line. Mythos 5 (Anthropic's higher-capability model, according to multiple reports) is tied to sensitive cybersecurity uses rather than ordinary consumer deployment. Dario Amodei (Anthropic co-founder and chief executive) is the company's public policy voice on AI risk. Howard Lutnick (U.S. commerce secretary in the Trump administration) is reported to have sent the export-control letter. The European Commission (EU executive based in Brussels) oversees the AI Act's general-purpose AI rules through the European AI Office. The EU AI Act (Regulation 2024/1689, in force since 2024) is the bloc's horizontal AI law.
How to read this story
The history
The dispute follows a broader shift in AI governance. The European Commission says the EU AI Act entered into force in 2024 and created specific duties for general-purpose AI models, with extra obligations for models designated as systemic risk. In the United States, President Joe Biden's 2023 AI executive order used federal reporting and safety-testing tools; President Donald Trump revoked it in January 2025 and later pursued a lighter national framework. The Anthropic order is different in tone: it uses export-control logic, closer to controls on chips or dual-use technology than to ordinary consumer software regulation.
The geopolitics
Frontier AI is increasingly treated as strategic infrastructure alongside chips, cloud capacity and cybersecurity tools. The U.S. order suggests Washington may use national-security powers to control model access, not only hardware exports. For Europe, that raises the familiar sovereignty issue: reliance on foreign platforms can become a geopolitical vulnerability when policy priorities diverge.
Why now
The trigger was the U.S. government's 12 June directive to Anthropic, which the company says cited national-security concerns and required it to block foreign-national access to Fable 5 and Mythos 5.
What to watch
Watch whether Anthropic announces restored access, whether the Commerce Department publishes or clarifies the licensing basis, and whether EU officials address the case under the AI Act's general-purpose AI framework. Any licence exemptions for allied countries would be especially relevant for Belgian and EU users.
Local impact
The most concrete Belgian impact is on the technology and cybersecurity sector: Belgian developers, university researchers and company security teams using Anthropic tools may need substitute models or revised workflows. The effect is not tied to one commune or region; it is concentrated in Belgium's digital businesses, research labs and public-sector IT teams.
International angle
This is a transatlantic technology-governance story. A U.S. order can remove access for EU users because the provider is American, while the European Commission is building its own AI Act regime for general-purpose models. The incident therefore exposes a gap between EU market rules and U.S. control over frontier-model infrastructure.
What this means for you
Belgian and EU teams using Anthropic models should check whether their workflows rely specifically on Fable 5 or Mythos 5, keep fallback models ready and review contracts for outage and regulatory-change clauses. The broader lesson is operational: frontier AI access can be interrupted by foreign security law with little warning.
What happens next
Anthropic says it wants to restore access as soon as possible, but any return likely depends on whether the U.S. government grants licences, narrows the directive or receives technical reassurance about the alleged jailbreak. EU policymakers may also look for whether the Commission or member-state authorities seek explanations from providers that offer systemic-risk models in the EU market.
Potential consequences
The shutdown could push European companies to diversify AI suppliers, keep fallback models available or negotiate stronger service-continuity terms. It could also strengthen political arguments for European frontier-model capacity, including firms such as Mistral AI. For regulators, the harder consequence is precedent: if model capability alone can trigger abrupt export controls, AI access may become less predictable for research, cybersecurity and business planning across Belgium and the EU.
Opposing perspectives
- Anthropic
Anthropic's statement argues that governments should be able to block unsafe deployments, but only through a transparent and technically grounded process. The company says this order did not identify a specific national-security basis, treated a limited jailbreak demonstration as decisive and forced a broad shutdown affecting lawful customers.
- U.S. national-security officials
The U.S. directive frames the models as sensitive strategic technology whose access by foreign nationals requires control. In that view, the cost of a sudden service interruption is outweighed by the risk that a high-capability model could be misused for cyber operations or other dual-use activity before regulators understand the exposure.
- European Commission AI Act framework
The Commission's AI Act framework treats the issue as one of lifecycle governance rather than a one-off access ban: general-purpose AI providers must document, evaluate and mitigate systemic risks. From that perspective, the episode reinforces why Brussels wants structured obligations, incident reporting and oversight capacity before frontier tools spread across the EU market.
Timeline
- 2024-08-01·The European Commission says the EU AI Act entered into force.
- 2025-04-21·Stelling et al. published research comparing industry safety practices with the draft EU general-purpose AI Code of Practice.
- 2026-05-17·David and Gervais published a benchmark paper on Mythos-linked bug rediscovery claims.
- 2026-06-12·Anthropic says it received the U.S. directive at 5:21 p.m. Eastern time.
- 2026-06-13·Anthropic's shutdown of Fable 5 and Mythos 5 was publicly reported across international outlets.
Glossary
- General-purpose AI model
- An AI model designed to perform a broad range of tasks and capable of being integrated into many downstream systems.
- Systemic risk model
- Under the EU AI Act framework, a high-impact general-purpose AI model subject to additional evaluation, mitigation and incident-reporting duties.
- Export controls
- Legal restrictions on the transfer of goods, software, technology or access to people or entities outside a controlled jurisdiction.
- Jailbreak
- A technique intended to bypass an AI model's safeguards so it produces outputs the provider tried to restrict.
Related to this story
Live connections from the Belgium Impulse ecosystem — not recommendations.
This briefing was prepared with AI assistance and reviewed by a Belgium Impulse editor before publication. methodology.