Anthropic's most capable publicly available AI model went dark on June 12, three days after launch. The US government ordered it offline over national security concerns. A jailbreak was involved, and Amazon played a role most people did not expect. Developers who had spent the week building on it woke up to find it gone, with no advance warning and no clear timeline for return. For 19 days, the most powerful AI model available to the public simply did not exist.
Anthropic launched Claude Fable 5 on June 9, 2026, as its first broadly available Mythos-class model and the most capable AI it had ever released to the general public. It shares underlying weights with Claude Mythos 5, a more restricted version limited to vetted partners through Project Glasswing, Anthropic's trusted cybersecurity consortium. The difference between them is safeguards. Fable 5 ships with classifiers that route high-risk cybersecurity, biology, chemistry, and distillation queries to Claude Opus 4.8 instead, triggering in fewer than 5% of sessions. Mythos 5 has fewer of those restrictions and significantly more raw capability. The capabilities were real. Stripe migrated a 50-million-line Ruby codebase in a single day using it. SWE-Bench Pro put it at 80.3%. The 1 million token context window and up to 128,000 token outputs made it genuinely different from anything previously available.
On June 12 at 5:21 pm ET, Commerce Secretary Howard Lutnick sent a letter to Anthropic CEO Dario Amodei invoking export control authority to suspend all access to Fable 5 and Mythos 5 for any foreign national, anywhere in the world, including Anthropic's own non-citizen employees. Anthropic had no way to verify nationality across hundreds of millions of users in real time. So it shut both models down for everyone. The trigger was a jailbreak. Amazon researchers had found a technique that prompted Fable 5 to read a codebase, identify software vulnerabilities, and in one case produce code demonstrating how a flaw could be exploited. White House AI adviser David Sacks described this publicly as a serious jailbreak of the model's guardrails. Anthropic disputed the severity directly. The vulnerabilities found were minor, already known, and reproducible by other publicly available models including GPT-5.5 without any jailbreak at all. The company argued the standard being applied would essentially halt all future frontier model launches across the entire industry.
Negotiations took roughly two and a half weeks. On June 26, Lutnick sent a second letter granting Mythos 5 access to approximately 100 US companies and federal agencies working on critical infrastructure defense. Fable 5 remained suspended while talks continued. On June 30, the Commerce Department lifted the export controls. Fable 5 redeployed globally on July 1 with a new cybersecurity classifier blocking the reported technique in over 99% of cases, per Anthropic and CAISI testing, with flagged requests rerouted to Opus 4.8. New commitments included expanded pre-release government testing access, a HackerOne bug bounty program for Fable 5 jailbreak submissions, a dedicated 24/7 jailbreak monitoring team, and participation in an interagency vulnerability clearinghouse alongside Amazon, Microsoft, and Google. Access returned for Pro, Max, Team, and select Enterprise users at up to 50% of weekly usage limits through July 7. After that, Fable 5 bills through usage credits at $10 per million input tokens and $50 per million output tokens. Mythos 5 remains restricted to vetted US organizations under Project Glasswing.
