Anthropic Offlines Fable 5 & Mythos 5: 10 Days After Trump’s AI Export Order

For AI practitioners and researchers, the sudden removal of Anthropic's Fable 5 and Mythos 5 represents a stark new reality: government export controls can now shutter access to frontier foundation models with almost no warning. This event shreds the assumption that AI releases will follow a predictable, transparent regulatory path, and it puts every lab on notice that model availability can be revoked overnight if national security concerns arise—even when the underlying directive is contested as opaque and unfounded.

Anthropic abruptly removed its two latest AI models, Fable 5 and Mythos 5, from public access on Friday, June 12, 2026, to comply with a new U.S. government export control directive that targets the dissemination of advanced AI capabilities to foreign nationals. The decision marks the most aggressive restriction on AI model availability by the U.S. government to date and directly follows an executive order signed by President Trump just 10 days earlier that established a voluntary framework for vetting AI models before their release.

“

For AI practitioners and researchers, the sudden removal of Anthropic's Fable 5 and Mythos 5 represents a stark new reality: government export controls can now shutter access to frontier foundation models with almost no warning.

Fable 5, released widely just days earlier, is described as a limited version of the more powerful Mythos 5, access to which Anthropic had already tightly restricted due to cybersecurity concerns. By taking both models offline, Anthropic effectively halted all external access—including to paying enterprise customers and research partners. The directive, which Anthropic says it received on Friday afternoon without advance notice, did not specify the precise national security concerns that triggered the action. The company strongly criticized the process, stating that while the government should have the authority to block unsafe deployments, such actions must follow a statutory process that is 'transparent, fair, clear, and grounded in technical facts.' Anthropic called the situation a 'misunderstanding' and expressed hope for a swift restoration of access.

The backdrop is the executive order signed by President Trump on approximately June 2, 2026, which empowers the federal government to review the most advanced AI systems for national security risks for up to one month prior to their public release. Critically, participation by AI developers was framed as voluntary, making the compulsory directive to Anthropic a significant and controversial escalation. It suggests the administration is willing to bypass the voluntary spirit of the order to enforce immediate restrictions, potentially under existing emergency export control authorities.

For the AI industry, this is a watershed event. Smaller AI labs will see it as a warning that even models developed in the U.S. can be summarily cut off if the government perceives a risk of foreign access. It blurs the line between hardware export controls—like those on advanced chips—and the intangible transfer of AI model weights and inference capabilities. The action raises profound questions about open-source AI, as many models are publicly released under permissive licenses. Anthropic’s models are proprietary, but the precedent could extend to any frontier system. Furthermore, the lack of clarity around the specific foreign national threat—whether it pertains to nation-state adversaries, end-user diversion, or broader cybersecurity risks—creates massive uncertainty for international collaboration, cloud API customers, and multinational corporations that rely on these tools.