On Fable and Export Controls
AI Village's statement on the US export control directive restricting access to Anthropic's Fable and Mythos models.
On June 12, 2026, the U.S. government took the drastic step of issuing an export control directive requiring Anthropic to suspend access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States. Anthropic says the practical result was that it had to disable both models for all customers to ensure compliance. On June 14, AI Village was proud to help draft and sign the open letter petitioning the government to reconsider its actions and return to a more grounded and transparent process for regulating both American models and AI companies. While we do believe nation-states have a right to regulate their companies and technologies, the current approach does far more harm than good to both the international and domestic cybersecurity environments.
Based on Anthropic’s public statement, the directive appears to have been prompted by concern over a narrow, non-universal jailbreak of Fable 5 and a demonstration Anthropic says involved identifying a small number of previously known, minor vulnerabilities. In the past, we have done multiple reports and events demonstrating the failures and weaknesses of guardrails, including two Generative Red Team events at DEF CON, one of which included participation from the U.S. Department of Defense. We understand Secretary Lutnick’s concerns that guardrails can be jailbroken by determined attackers, and that their long-term protection for a model is limited. However, this is not a problem exclusive to Fable 5 and Mythos 5, or even Anthropic as a whole; instead, it is a fundamental limitation of LLMs as a technology, similar to their problems with hallucinations. A policy standard that requires perfect jailbreak resistance would be impossible to apply consistently across frontier models.
That distinction matters. A model’s ability to inspect code and identify or fix software flaws is not inherently offensive. It is also one of the core capabilities defenders need when maintaining real systems under shrinking patch windows.
Additionally, the act of removing access to two of the most powerful coding models does more to harm the cybersecurity of U.S. persons and infrastructure than it does to protect them. For quite a while now, the time between a flaw in a software library being reported and that flaw being exploited has steadily been shrinking, and AI coding agents have become a key way for many independent and open-source developers to keep up with the rising tide and ensure their users are safe. Many of these developers are not U.S. citizens, and the practical shutdown of Fable 5 and Mythos 5 also affected U.S. users. By removing these tools from defenders, the U.S. government is actively harming the overall security of the global software ecosystem.
We hope that Secretary Lutnick listens to members of the AI and cybersecurity communities and reconsiders the current directive. The administration has already recognized, in its June 2 Executive Order on advanced AI innovation and security, that AI-enabled vulnerability discovery can strengthen the defense of federal systems, critical infrastructure, and vulnerable local institutions. That same principle should guide export-control decisions: policy should be technically grounded, transparent, and developed with input from the AI and cybersecurity communities before defensive tools are removed from the people using them to secure real systems.
America leads not only in the development of AI but also in the responsible and ethical use of it. We hope that the U.S. government continues to lead in this area, and that it does so with a more open and transparent process that includes input from the AI and cybersecurity communities.