Anthropic Pulls Claude Fable 5 and Mythos 5 After US Government National Security Order

Government Shutdown Order Escalates Anthropic Conflict

Anthropic has disabled Claude Fable 5 and Mythos 5, two AI models launched three days earlier, in response to a US government directive received on Friday evening. According to Wired AI, the order cited national security concerns and cited evidence of a jailbreak technique that could circumvent the models’ safety restrictions preventing responses on cybersecurity, biology, and chemistry topics. The export control directive required Anthropic to revoke access for all foreign nationals, including non-US employees, but the company chose to take the models offline entirely for all users rather than implement selective restrictions.

The Jailbreak at the Center of the Dispute

The technical disagreement hinges on the severity of the vulnerability the government identified. According to Wired AI, Anthropic’s interpretation is that the government discovered a narrow, non-universal bypass technique involving instructing the model to read specific code and identify flaws. The company maintains that these vulnerabilities are minor, previously known, and discoverable in other publicly available models without requiring any jailbreak method. Anthropic has not disclosed how long the models will remain offline or whether it plans to challenge the order through the courts.

Backdrop: Months of Escalating Tensions

This enforcement action represents a dramatic intensification of the Trump administration’s confrontation with Anthropic. Earlier in 2026, the Department of Defense classified Anthropic as a “supply chain risk” after the AI maker attempted to establish limits on military applications of its technology. That designation barred federal agencies and contractors from deploying Anthropic’s models. Anthropic responded by filing lawsuits against the administration in response to the supply-chain designation, setting the stage for Friday’s government order.

Anthropic’s Compliance Strategy

Rather than implement the technically feasible partial restriction—blocking foreign nationals while allowing US-based users and employees—Anthropic chose complete suspension. According to Wired AI, the company justified this approach in a blog post by stating that the government’s letter did not provide written technical specifications of the vulnerability, only a demonstration of what it described as a simple jailbreak technique. By taking both models offline globally, Anthropic appears to signal either uncertainty about implementing a compliant partial restriction or a strategic decision to avoid operating under government-imposed safeguard requirements.

Why This Matters

This incident sets a precedent for how the US government will enforce export controls against frontier AI developers during a period of heightened national security scrutiny. If the government can order the complete suspension of commercially available models based on verbal evidence of jailbreak vulnerabilities without written technical justification, AI developers face pressure to either accept government safety requirements or risk market disruption. For customers relying on Claude Fable 5 for legitimate cybersecurity testing, the shutdown creates immediate operational disruption and raises questions about whether similar controls will be imposed on competing models offering comparable capabilities.