Calvin Wankhede / Android Authority
TL;DR
- The US government issued an emergency export control directive forcing Anthropic to immediately suspend access to its flagship Fable 5 and Mythos 5 AI models for all foreign nationals due to national security concerns.
- To guarantee absolute compliance with the order, Anthropic had to completely disable both models for all customers worldwide, including domestic US citizens.
- The restriction stems from a codebase “jailbreak” exploit that allows users to use these models to scan for software vulnerabilities.
- Anthropic strongly pushes back, arguing the flaw is minor, present in competitor models like GPT-5.5, and shouldn’t trigger a commercial blackout.
In a stunning and unprecedented move, the US government has issued an emergency export control directive forcing Anthropic to immediately suspend all access to its flagship Fable 5 and Mythos 5 artificial intelligence models within Claude.
According to an official statement released by Anthropic, the US government’s directive arrived abruptly on June 12, citing national security authorities but offering little initial documentation. The US government has directed Anthropic to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees.
However, the net effect of this order is that Anthropic has had to disable Fable 5 and Mythos 5 for all its customers, including US nationals, to ensure complete compliance. Anthropic notes that access to all other Anthropics models is not affected.
While Anthropic has complied with the legal mandate, the company expressed strong disagreement with the decision, calling it a “misunderstanding” and warning that such an opaque intervention could freeze progress across the entire tech sector.
Don’t want to miss the best from Android Authority?
The core of the issue appears to center on a discovered “jailbreak” — a technique used to bypass an AI model’s built-in safety guardrails. Specifically, the government reportedly became aware of a narrow, non-universal jailbreak where a user could trick the model into scanning a specific codebase to find and patch software flaws.
Anthropic fiercely defended its security posture, noting that it had logged thousands of hours red-teaming Fable 5 with the US government and the UK AI Safety Institute before its launch. The company argues that the vulnerabilities exposed by this exploit are relatively minor and can already be achieved using other publicly available frontier models, including OpenAI’s GPT-5.5. Furthermore, Anthropic emphasized that perfect jailbreak resistance is a technical impossibility at present, advocating instead for a “defense in depth” strategy paired with stringent monitoring.
If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers. Anthropic statement
The heavy-handed nature of this restriction draws immediate parallels to the Department of Commerce’s placement of HUAWEI on the Entity List in 2019. Back then, the US government used export controls under the banner of national security to completely sever a foreign hardware giant from US supply chains and software. However, while the HUAWEI ban targeted a foreign entity to protect domestic interests, this new enforcement effectively sees the US government recall its own homegrown software assets out of geopolitical concerns.
If this aggressive move is applied evenly across the AI industry, any discovered software vulnerability could suddenly result in a commercial blackout for American AI companies. Obviously, that’s not good for customers, nor is it an acceptable business risk for AI companies themselves, who have poured billions of dollars into research and development (and made our phones and other tech more expensive in this pursuit).
Anthropic says it is actively working with officials to resolve the matter and restore access.
Thank you for being part of our community. Read our Comment Policy before posting.
