Anthropic Restores Claude Fable 5 After U.S. Lifts Emergency Export Controls

Anthropic is bringing Claude Fable 5 back online worldwide after the U.S. Commerce Department reversed the emergency export controls that forced the model offline in mid‑June.

The restrictions, imposed on June 12, required Anthropic to block access for all foreign nationals—including its own non‑citizen employees—leaving the company no practical way to keep the model available. As a result, both Fable 5 and its more tightly governed sibling Mythos 5 were shut down globally.

Fable 5 returns on July 1 across Claude.ai, the Claude Platform, Claude Code, and Claude Cowork. Mythos 5, which carries fewer guardrails, regained limited access on June 26 for roughly 100 U.S. critical‑infrastructure organizations, with broader availability still under negotiation.

The trigger for the shutdown was a jailbreak discovered by Amazon researchers. The prompt caused Fable 5 to identify a handful of software flaws and, in one case, generate code demonstrating how a vulnerability could be exploited. Anthropic downplayed the severity, arguing that similar behavior appears in weaker models such as Claude Opus 4.8, OpenAI’s GPT‑5.5, and China’s Kimi K2.7, and characterizing the output as routine defensive security work rather than evidence of a dangerous capability.

Still, federal officials and Amazon viewed the finding as serious enough to justify emergency controls. To address the concern, Anthropic trained a new classifier designed to detect and block the specific jailbreak technique. The company says the filter now stops the method in more than 99% of attempts, redirecting blocked queries to Opus 4.8 and notifying users. The trade‑off is a higher rate of false positives in normal coding and debugging tasks.

Commerce Secretary Howard Lutnick said the department spent two weeks reviewing the models with Anthropic before lifting the restrictions. The company agreed to proactively search for security issues, coordinate future launches with regulators, and report malicious use. Negotiations were reportedly led by co‑founder Tom Brown rather than CEO Dario Amodei, who has clashed with the administration throughout the year.

The episode exposed tensions across the industry. Reports indicated Amazon’s concerns helped drive the initial shutdown, while critics argued the government overreacted. Researchers noted that the pause arrived just as low‑cost Chinese open‑source models were gaining momentum, raising fears that halting U.S. systems would give competitors time to catch up.

Anthropic is now pushing for a standardized way to score jailbreak severity, working with Amazon, Microsoft, Google, and others. The proposed framework evaluates capability gain, breadth of unlocked attacks, ease of weaponization, and discoverability. For the most severe cases—such as jailbreaks enabling attacks on critical infrastructure—Anthropic says it will deploy fixes immediately and maintain round‑the‑clock monitoring.

The company also launched a HackerOne program for reporting new Fable 5 jailbreaks and promised the U.S. government earlier access to test future frontier models. The broader concern remains unresolved: a June 2 executive order created a voluntary review path for advanced models but no mandatory process. Fable 5 never went through that channel, leaving regulators to rely on improvised export controls when acting quickly.