📰 Key Takeaways

Trump’s administration shows signs of softening its stance on Anthropic’s cybersecurity models Mythos 5 and Fable 5, which were banned two weeks ago. Commerce Secretary Howard Lutnick sent a letter to Anthropic’s Chief Operating Officer Tom Brown on Friday, confirming that necessary safeguards are in place, allowing over 100 designated US government agencies and companies to access Mythos 5. The administration also expanded authorization to non-US employees within these organizations, with Anthropic’s own non-US staff also included in this lifting of the ban (the original ban had excluded them all).

The original ban stemmed from security researchers claiming they could easily bypass the guardrails of both models, forcing Anthropic to take both offline. The new government directive only targets Mythos 5, with no explanation regarding Fable 5’s return. Fable 5 was released just days before the ban took effect, touting more robust guardrails, but ultimately met the same fate.

Anthropic confirmed the progress publicly on X on Friday, stating that Mythos 5 will be prioritized for redeployment to US organizations responsible for “operating and defending critical infrastructure,” and will continue negotiating with the government to further expand access to Mythos 5 while also seeking to restore general public use of Fable 5.

💬 JudyAI Lab’s Perspective

The Trump administration’s easing of restrictions on Anthropic’s two models shows that Mythos 5’s partial lift reveals a critical insight: between security vulnerabilities and regulatory intervention, AI access can be revoked in an instant—a warning for organizations relying on specific models.

This incident highlights an emerging structure in the AI industry: when security researchers publicly point out guardrail flaws, even a company of Anthropic’s scale must pull everything offline. What’s even more noteworthy is the government’s prioritization logic in lifting the ban—restoring access first to organizations involved in “operating and defending critical infrastructure,” while Fable 5 has received no explanation whatsoever. This means AI access is no longer purely a business matter—compliance capabilities, risk response processes, and even channels for government negotiation will determine how quickly a product can recover after a crisis. Fable 5 was released just days before the ban took effect and still met the same fate, showing that PR timing is not an effective buffer during regulatory crises.

If your product relies on a single cloud AI model, you should now map out backup options—don’t wait until regulators come knocking to plan alternatives.

📅 Source Information

🔗 Further Reading