In a dramatic turn of events, the U.S. government delivered a significant blow to AI developer Anthropic last Friday, ordering the immediate cessation of access to two of its most powerful artificial intelligence models. Citing critical national security concerns, the directive targets Claude Fable 5 and Claude Mythos 5, forcing a global shutdown of these advanced systems.
Anthropic confirmed its compliance via an announcement on X, though the company explicitly stated its disagreement with the government’s decision. The order, received at 5:21 pm ET, mandates that both models be disabled for all users worldwide, extending beyond the foreign nationals that an export control order typically addresses.
A Sudden Shutdown: National Security Concerns Halt Advanced AI
The government’s directive is officially framed as an export control action, aiming to restrict access to these AI models by foreign nationals. However, in a detailed blog post, Anthropic shared its understanding that the core concern revolves around a claimed “jailbreak” of its Fable 5 model.
According to Anthropic, the government has so far provided only verbal evidence of a “potential narrow, non-universal jailbreak.” This alleged exploit reportedly involves prompting the model to analyze specific codebases and identify software flaws, a capability Anthropic argues is already prevalent in other publicly available models.
Unpacking the Models: Mythos, Fable, and the Hunt for Vulnerabilities
The impact of this shutdown is particularly significant due to the capabilities of the models involved. Claude Mythos 5 stands as Anthropic’s most advanced AI, initially previewed in early April and kept under strict wraps precisely because of its extraordinary aptitude for uncovering security vulnerabilities in software.
Anthropic stated that Mythos identified critical flaws across every major operating system and web browser it was tested against. Rather than a broad release, the company initiated “Project Glasswing,” a controlled program that shared Mythos 5 with approximately 50 meticulously vetted organizations for defensive cybersecurity work. These partners included tech giants like Amazon, Apple, Google, Microsoft, and the cybersecurity firm CrowdStrike.
Claude Fable 5, launched just three days before the government order, was Anthropic’s answer to the commercial demand for such a powerful AI. It was essentially a version of Mythos equipped with stringent guardrails designed to prevent responses in high-risk domains like cybersecurity and biology.
Anthropic positioned Fable 5 as safe enough for general public release, and benchmark tests from Vals AI, an AI performance tracking company, quickly labeled it the most capable AI model available to the public. Its sudden removal leaves a notable void in the publicly accessible frontier of AI technology.
The “Jailbreak” Controversy: Anthropic’s Defense Against Government Claims
Anthropic vigorously contests the severity and implications of the alleged Fable 5 jailbreak. The company emphasizes that the specific capability described—identifying software flaws from a codebase—is routinely utilized by cybersecurity professionals for purely defensive purposes.
Furthermore, Anthropic points out that a similar “level of capability” is widely accessible in other public models, specifically naming OpenAI’s GPT-5.5. The company also clarified that its robust safety mechanisms, including independent classifier systems, operate separately from the core model.
This means that even if a user could somehow prompt Fable 5 to bypass an initial refusal, the underlying protections against generating genuinely dangerous outputs would theoretically remain active. Despite these reassurances, the government moved forward with its unprecedented directive.
The Ironic Twist: Safety Warnings and Regulatory Scrutiny
Anthropic has not hidden its frustration, stating, “We disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people.” The company warns that applying such a stringent standard across the industry could effectively halt all new model deployments by frontier AI providers.
The situation presents a stark irony for Anthropic, a company that has consciously built its public image on being a safety-first alternative to its rivals. Its deliberate caution in restricting Mythos 5, promoting it as a model too powerful and potentially dangerous for public release, seems to have inadvertently attracted the very government scrutiny it sought to mitigate.
Many observers note this development could significantly impact Anthropic, particularly as it heads toward a widely anticipated IPO this year. OpenAI CEO Sam Altman, a frequent critic of Anthropic’s approach, previously characterized their handling of Mythos as “fear-based marketing.”
In an April podcast, Altman likened it to building a “bomb” and then selling “bomb shelters” for a premium. While he didn’t predict a government shutdown, Altman shrewdly identified that when a company spends months highlighting the unique dangers of its AI, the world—including regulatory bodies—tends to pay very close attention.
Source: TechCrunch – AI
