Anthropic Shuts Down Fable 5 and Mythos 5 Worldwide After US Government Export Control Directive
Anthropic disabled its two most advanced AI models globally after a U.S. government directive ordered blocking access for all foreign nationals, citing a potential jailbreak technique.
Anthropic has disabled its two most capable AI models, Fable 5 and Mythos 5, after the U.S. government issued an export control directive late on June 12 ordering the company to block access for any foreign national, whether inside or outside the United States, including Anthropic's own foreign-national employees. Because the company says it cannot reliably separate foreign users from the rest of its base in real time, the practical result is a worldwide shutoff of both models. All other Anthropic models remain online.
The directive, delivered at 5:21 PM ET on Friday, June 12, 2026, mandates that no foreign national be permitted access to the two models. Anthropic publicly acknowledged the order via a statement on X (formerly Twitter), noting that selectively enforcing nationality-based access controls at scale is operationally impractical, so the company extended the restriction to its entire global customer base to ensure full legal compliance.
The government's concern centers on an alleged jailbreak technique that involves instructing the model to read a specific codebase and identify software flaws. According to Anthropic, the government has so far only provided verbal evidence of a potential narrow, non-universal jailbreak. Anthropic reviewed a demonstration of the specific technique and found that it surfaced only a small number of previously known, minor vulnerabilities. Critically, the company states that other publicly available models, including OpenAI's GPT-5.5, can produce the same output without requiring any bypass.
Anthropic has publicly acknowledged that perfect jailbreak resistance is not achievable by any model provider today. In response, the company adopted a defense-in-depth strategy: engineering non-universal jailbreaks to be narrow in scope and universal jailbreaks to be prohibitively expensive to produce, combined with comprehensive monitoring to detect and shut down successful attacks rapidly. As part of this strategy, Anthropic implemented a 30-day customer data retention policy for Mythos-class models to support ongoing jailbreak research and mitigation.
The shutdown comes just days after researcher 'Pliny the Liberator' demonstrated a jailbreak of Claude Fable 5, using multi-agent decomposition and Unicode tricks to bypass safety classifiers and generate exploit code. That incident, reported on June 10, highlighted the ongoing cat-and-mouse game between AI safety measures and adversarial techniques. The government's directive appears to be a direct response to such vulnerabilities, though Anthropic argues the specific technique in question is neither novel nor unique to its models.
Anthropic has apologized to affected customers and stated it is working urgently to restore access, with a commitment to release further technical details within 24 hours of the directive. The incident underscores the growing tension between AI capability advancement and national security concerns, as governments increasingly scrutinize frontier models for potential dual-use risks. The global shutdown of Fable 5 and Mythos 5 marks one of the most aggressive government interventions in AI deployment to date, setting a precedent that could reshape how advanced AI systems are regulated and distributed internationally.
The CyberScoop report adds that the Commerce Department's directive, signed by Secretary Howard Lutnick, also restricts foreign-born Anthropic employees from accessing the models, forcing the company to disable them globally. Anthropic disputed the severity of the alleged jailbreak, calling it a 'narrow, non-universal' technique that involved prompting the model to read code and identify software flaws — capabilities it says are already present in OpenAI's GPT-5.5. The article also notes that Amazon may have flagged the issue, and that the NSA had been given access to Mythos 5 for offensive cyber operations prior to the ban.
Anthropic's statement, published after the initial shutdown announcement, clarifies that the government's jailbreak demonstration only uncovered a small number of previously known, minor vulnerabilities that other publicly available models can also find. The company notes that Mythos 5 is particularly effective at identifying long-standing bugs in legacy critical infrastructure systems, a dual-use capability that makes it valuable for defense hardening but catastrophic in adversarial hands. Despite the global disablement, older Claude variants remain available, and Anthropic argues the removal only modestly raises the bar for less-resourced cybercriminals, as other models on the black market may be equally effective.
The Commerce Department's Bureau of Industry and Security (BIS) issued the directive under the Export Control Reform Act, marking the first time this ECRA provision has been applied to a commercially available AI model. Anthropic received the order on June 12 and immediately disabled global access to both models. The action was reportedly triggered by a jailbreak vulnerability in Fable 5 that could expose Mythos 5's advanced cybersecurity capabilities, though Anthropic downplays the risk. Over 80 cybersecurity executives, including leaders from Nvidia and Adobe, have signed an open letter urging the administration to lift the restrictions.
On June 27, 2026, the U.S. government notified Anthropic that Claude Mythos 5 can be redeployed to a defined set of domestic critical infrastructure organizations in sectors such as energy and healthcare. Anthropic stated it is moving quickly to restore access for these validated entities, marking the first official restoration since global suspension began on June 12. The phased return maintains strict controls while working toward broader expansion and the eventual restoration of Claude Fable 5 for unrestricted use.