Washington Eases Restrictions on Anthropic’s Claude Mythos 5 After National Security Standed-Off

The Sudden Blackout of Claude Fifth Generation Models

On June 9, 2026, Anthropic launched Claude Fable 5 and Claude Mythos 5, representing its most advanced artificial intelligence systems to date. Priced at 10 dollars per million input tokens and 50 dollars per million output tokens, these models were introduced at less than half the cost of the Claude Mythos Preview. Fable 5 was designed for general use, boasting state-of-the-art performance on benchmarks for software engineering, vision, and scientific research. To mitigate risks, Anthropic equipped Fable 5 with conservative safeguards that routed sensitive queries to Claude Opus 4.8, a backup system that triggered in fewer than 5 percent of user sessions. Meanwhile, Mythos 5 was launched as a specialized tool with lifted safeguards, specifically tailored for cyberdefense and deployed through Project Glasswing in partnership with the United States government.

The rollout was short-lived. Just three days later, on June 12, 2026, Anthropic abruptly suspended global access to both models. The sudden shutdown disrupted developers, researchers, and organizations running complex simulations and AI agents, leaving the tech industry searching for answers as the company worked behind the scenes to address the disruption.

National Security Fears and Export Controls

The disruption stemmed from a strict export control directive issued by the United States government under the Trump administration. The order prohibited Anthropic from allowing foreign nationals, whether located inside or outside the United States, to access Fable 5 and Mythos 5. Because separating foreign users from domestic ones instantly was technically unfeasible, Anthropic was forced to disable both models globally to ensure immediate compliance.

Government officials cited national security concerns, pointing to a potential vulnerability that could allow users to bypass the safety guardrails, or jailbreak, Fable 5. The fear was that malicious actors could exploit the advanced cybersecurity capabilities of these frontier systems to cause severe damage. Anthropic initially pushed back against the recall, arguing in a blog post that the vulnerability was narrow rather than universal. The company also pointed out that similar capabilities were already accessible in other publicly available models, questioning whether a commercial recall of this scale was justified.

A Partial Restoration for Critical Infrastructure

Two weeks of intensive negotiations between Anthropic and federal regulators yielded a breakthrough. On June 26, 2026, Commerce Secretary Howard Lutnick sent a letter to Anthropic chief compute officer confirming that the company had made significant progress in addressing the security risks. Lutnick granted permission to redeploy the Mythos 5 model to a select group of roughly 100 trusted partners, federal agencies, and organizations that operate and defend critical infrastructure.

While this clearance eases the immediate confrontation, the general-use Fable 5 model remains offline. Commerce Department spokesman Benno Kass emphasized that the rapid resolution balanced national security with the goal of keeping the nation at the forefront of global AI leadership. Anthropic is currently working to provision access for the approved providers while continuing discussions with the government to eventually restore Fable 5 for general public use. To prevent future abrupt shutdowns, the company is also collaborating with federal authorities to establish a permanent policy framework for managing security risks in frontier AI models.

This swift regulatory intervention and subsequent compromise establish a tense precedent where the deployment of frontier AI is no longer solely a corporate decision, but one subject to immediate state oversight and national security vetoes.