The US government issued Anthropic an export control directive: starting Friday, June 13, 2026, the company must block access to Claude Fable 5 and Mythos 5 for all foreign nationals — whether residing outside the US or living within it. The directive also covered Anthropic employees who are not US citizens. The practical outcome was sweeping: the company disabled both models for all customers globally to ensure full compliance.
Key takeaways
- The export control directive reached Anthropic on June 13, 2026 at 5:21pm ET
- The government cited national security concerns and the risk of jailbreaking Fable 5
- Anthropic disabled Fable 5 and Mythos 5 for all customers — no other models are affected
- The company disputes the decision: disclosed exploits are non-universal and achievable via other models, including GPT-5.5
- Anthropic committed to publishing technical details within 24 hours of its statement
The directive and Anthropic's response
The directive, published by Anthropic in an official statement, provided no specific technical justification according to the company. The government communicated that it had become aware of a technique to bypass — "jailbreak" — Fable 5's safety measures. Anthropic acknowledged reviewing a demonstration of this technique but assessed the discovered vulnerabilities as minor and reproducible in other publicly available models.
In its official statement, Anthropic described the only evidence the government provided as a jailbreak based on asking the model to read a specific codebase and fix software flaws. The company assessed that the level of capability demonstrated was achievable with generally available models — including OpenAI's GPT-5.5 — and is used daily by defensive security practitioners.
We have not even received a disclosure of a concerning non-universal potential jailbreak that led to a harmful result. The potential jailbreaks that have been disclosed to us are either entirely benign responses or are minor findings that provide no Mythos-specific uplift.
The company noted that Fable 5 was tested for weeks before launch by US and UK governments, external organizations, and internal red teams over thousands of hours in total. No tester succeeded in finding a "universal jailbreak" — a method capable of broadly bypassing safety measures across diverse contexts simultaneously.
Context: Fable 5 safeguards and the "defense in depth" strategy
Fable 5 and Mythos 5 are from the so-called Mythos class — Anthropic's most powerful models to date, surpassing earlier Claude versions on complex reasoning and agentic tasks. At launch, Anthropic implemented restrictive safeguards limiting access to cybersecurity, biology, and chemistry-related information — a move that drew criticism from some researchers as excessively broad.
Anthropic's defensive strategy was built on the premise that perfect jailbreak resistance is not achievable with current technology. The company therefore opted for a "defense in depth" approach: making jailbreaks either narrow in scope or extremely expensive to execute, while maintaining intensive monitoring of exploitation attempts. The 30-day customer data retention policy — a decision that carried real business costs for Anthropic — was part of this strategy, enabling faster detection and response to threats.
In the days prior to this directive, Anthropic was already at the center of controversy over Fable 5, having to apologize for invisible restrictions that blocked responses to basic biology questions without informing users. Now the US government has escalated in the opposite direction: the concern is no longer that the filters are too restrictive, but that they are allegedly insufficient.
Market consequences and regulatory precedent
The US administration's decision raises fundamental questions about the limits of government intervention in commercial AI deployments. Anthropic stated directly in its release that if the same standard — blocking a model after a narrow, non-universal jailbreak is found — were applied consistently across the industry, it would effectively halt all new frontier model deployments across all providers.
At the same time, the company stated its support for the government's ability to block unsafe AI deployments, with a critical caveat: such a process should be transparent, grounded in technical facts, and procedurally fair. The current decision, Anthropic argues, meets none of these criteria.
Access to all other Anthropic models — including Claude Opus, Sonnet, and Haiku — remains unaffected. Fable 5 and Mythos 5 disappear from the offering for all customers regardless of location or citizenship. The shutdown was not limited to foreign users, because complying with such a partial restriction would require verifying the citizenship of every user in real time — a technically impractical task in such a short timeframe.
Why this matters
This is the first confirmed instance of the US government applying an export control instrument against a commercial AI model and forcing its immediate shutdown for all global users. The operational scale of this precedent is unprecedented: Fable 5 and Mythos 5 had been available to hundreds of millions of users.
If the US administration maintains this approach for future model generations, every new frontier model release will face the risk of sudden takedown without a transparent technical review process. For companies building products and services on Anthropic models, this creates new operational risk that was previously not priced into vendor lock-in?vendor lock-in: Vendor lock-in is a situation where switching to a competing provider’s product or service is so costly or technically difficult that the customer becomes tied to a single supplier. and provider diversification strategies. The question also opens whether other jurisdictions — the EU, China, India — will adopt similar control instruments against AI models, potentially fragmenting global access to AI tools along political lines.
What comes next
- Anthropic committed to publishing detailed technical data on the disclosed jailbreaks within 24 hours of its June 13 statement
- The company is working toward restoring model access and frames the current block as a misunderstanding it intends to resolve with the government
- The outcome of potential legal or administrative action will be a key factor to watch: Anthropic publicly stated the decision fails transparency and procedural fairness standards, which may be a preliminary step toward challenging the directive
