Anthropic, the AI safety company behind the Claude family of models, was ordered by the US government on national security grounds to immediately suspend all access to its Fable 5 and Mythos 5 models for any foreigner, whether located inside or outside the United States. The directive, received on July 14, 2025, at 5:21 PM Eastern Time, forced Anthropic to disable the two models for its entire global customer base to ensure full compliance, while leaving access to all other Claude models unaffected.
What Happened: A Government Directive With Limited Explanation
Anthropic disclosed the situation publicly through a formal statement on its website and across its official social media channels. According to the company, the directive arrived with no specific national security justification included in the letter. The government communicated verbally its belief that a method of bypassing, or “jailbreaking,” Fable 5 had been identified, raising concerns about potential misuse of the model’s capabilities.
The directive was issued citing national security export control authorities, a legal framework typically used to restrict the transfer of sensitive technologies to foreign entities. By applying it to AI model access, the government effectively treated Fable 5 and Mythos 5 as controlled technologies, requiring Anthropic to cut off any foreign national from using them, including its own employees who hold foreign citizenship.
Anthropic’s Response: Disagreement and Transparency
While confirming compliance with the order, Anthropic pushed back publicly against the reasoning behind it. The company stated that it reviewed the specific jailbreak technique referenced by the government and found it to be limited in scope and impact. According to Anthropic’s statement, the demonstrated technique uncovered only a small number of previously known, minor vulnerabilities, all of which were described as relatively simple and discoverable using other publicly available AI models without requiring any bypass of Fable 5.
Anthropic also stressed that it has not received a single disclosure of a jailbreak that produced a harmful result. The company characterized the potential jailbreaks shared with it as either generating entirely benign responses or producing minor findings that offer no specific uplift unique to Mythos 5.
The Safety Architecture Behind Fable 5
Extensive Pre-Launch Red-Teaming
Anthropic outlined in its public statement the significant safety work that preceded Fable 5’s launch. In the weeks before release, the company collaborated with the US government, the UK AI Safety Institute (AISI), multiple private third-party organizations, and internal teams to conduct thousands of hours of red-teaming exercises focused on testing Fable’s safeguards. Those tests concluded that Fable 5’s protections were substantially more effective than those of any previously deployed model in the industry.
No Universal Jailbreak Found
A key point in Anthropic’s defense is that no tester, during or after the pre-launch evaluation period, has succeeded in finding a universal jailbreak, defined as a method capable of broadly bypassing the model’s safeguards to unlock a wide range of potentially harmful capabilities. The company acknowledged that non-universal jailbreaks, which can elicit specific information under particular circumstances, are a known and unavoidable reality across the entire AI industry.
A Defense-in-Depth Strategy
Rather than claiming perfect jailbreak resistance, Anthropic adopted what it describes as a defense-in-depth approach with Fable 5. This strategy is designed to ensure that any successful jailbreaks remain either narrow in scope or extremely costly to execute. It is paired with continuous monitoring systems intended to detect and shut down attacks quickly after they occur.
As part of this strategy, Anthropic introduced a 30-day customer data retention policy specifically for Fable, a decision the company acknowledged carries commercial costs. The policy was implemented to allow ongoing research into jailbreak attempts and enable faster mitigation of any successful exploits.
Immediate Impact on Users and Developers
For users of Anthropic’s consumer and developer products, the suspension means that new sessions previously running on Fable 5 will be automatically redirected to either a user-selected default model or to Opus 4.8. Anthropic confirmed through its developer-facing ClaudeDevs channel that all other Claude models remain fully available.
The disruption is particularly notable given that Fable 5 appears to have been positioned as one of Anthropic’s flagship offerings. Users had reportedly noted that its safety filters were, in some cases, considered overly broad, a reflection of the aggressive safeguard settings Anthropic had implemented to reduce misuse risk in cybersecurity-related tasks.
Broader Implications for AI Governance and Export Controls
A Precedent-Setting Move
The US government’s directive represents a significant moment in the evolving relationship between federal authorities and AI developers. Applying export control mechanisms to access to a specific AI model, rather than to hardware or underlying weights, is a relatively novel use of national security law. It raises immediate questions about how authorities will define and regulate advanced AI capabilities going forward.
Uncertainty Over Evidence
Anthropic’s statement noted that, as of publication, the government had provided only verbal evidence of the concerning jailbreak technique. The lack of formal written disclosure has placed the company in the position of complying with a directive it believes is not justified by the available evidence, while simultaneously being unable to resolve the underlying dispute through a direct technical review of the government’s full findings.
The company expressed a willingness to engage further with authorities and indicated it would continue to advocate for a transparent and evidence-based process. The case is likely to draw close attention from other AI developers, policymakers, and civil liberties advocates as questions about the scope of government authority over commercially deployed AI systems continue to grow.
- Fable 5 and Mythos 5 access has been suspended globally for all foreign nationals, including Anthropic staff.
- All other Claude models, including Opus 4.8, remain fully operational and unaffected by the directive.
Anthropic has stated it will provide further updates as the situation develops. The full statement is available on the company’s official website at anthropic.com.
Source : https://www.anthropic.com/news/fable-mythos-access
The US government, citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees.
— Anthropic (@AnthropicAI) June 13, 2026
The net effect of…
Regis Vansnick is a recognized expert with extensive experience at the intersection of technology, business, and innovation. His professional career is marked by a deep understanding of digital transformation and strategic management.
