U.S. Orders Anthropic to Suspend Fable 5 and Mythos 5 Access for Foreign Nationals

Anthropic announced it will "abruptly disable" its advanced AI models, Claude Fable 5 and Mythos 5, following a U.S. government directive to suspend access for foreign nationals due to national security concerns. The company received the order late in the evening and is working to address what it believes is a misunderstanding to restore access. This suspension affects only these specific models, with other models remaining unaffected by the directive. The government reportedly believes it has discovered a method to bypass or "jailbreak" Fable 5, which could potentially expose vulnerabilities. Anthropic acknowledged a demonstration of this technique, which identified minor vulnerabilities that are relatively simple and can be discovered by other publicly available models without needing a bypass. Despite these claims, the company maintains that its models are secure and that the vulnerabilities are not unique to its technology. The suspension comes shortly after the launch of Claude Fable 5 and Mythos 5, the latter being a version with fewer restrictions in cybersecurity applications. Mythos 5, described as having unparalleled cybersecurity capabilities, remains accessible to a select group of vetted cyber defenders and critical infrastructure operators. Anthropic has emphasized the strong guardrails it has in place to prevent misuse, particularly in cybersecurity contexts. Anthropic has implemented safety classifiers designed to detect and block potential misuse, including jailbreak attempts. These classifiers prevent the main model from responding to harmful requests related to cyber attacks or exploit development. The company notes that Mythos-class models are adept at identifying and exploiting software vulnerabilities, which could give attackers a strategic edge. The company highlighted the capabilities of its Mythos-class model, which can rapidly convert newly disclosed software vulnerabilities into working exploits, reducing the time from weeks to mere hours or minutes. This capability poses a challenge to traditional software patching practices, which may no longer be effective against such rapid exploitation. In response to cybersecurity queries, Fable 5's protections redirect users to Claude Opus 4.8, another capable model. Anthropic argues that no universal jailbreak methods have been developed against its latest models, and its safeguards have proven more effective than previous models. However, it acknowledges that perfect jailbreak resistance is unattainable, as all industry safeguards are vulnerable to non-universal jailbreaks. Anthropic expressed concerns over the government's directive, stating that the evidence provided was verbal and based on a narrow, non-universal jailbreak. The company reviewed a report believed to be the basis for the directive and found that similar capabilities are available in other models, such as OpenAI's GPT-5.5, which are used by defenders to secure systems. The company advocates for a transparent, fair, and technically grounded statutory process for addressing AI deployment concerns. Earlier this year, the U.S. Department of Defense labeled Anthropic a "supply chain risk" after the company opposed military use of its technology, leading to two lawsuits filed by Anthropic to challenge the designation.