In a dramatic and unprecedented move, Anthropic recently announced the immediate withdrawal of its highly anticipated Claude Fable 5 and Mythos 5 models. Overnight, these advanced AI tools became inaccessible to all users worldwide. This abrupt decision stemmed directly from a US government directive, citing critical national security concerns.
Anthropic confirmed a “US government, citing national security authorities,” issued an export order. This mandate specifically required the company to disable access for “any foreign national, whether inside or outside the United States, including foreign national Anthropic employees.” To ensure full compliance, Anthropic took the broader step of disabling both models for everyone, both domestically and internationally.
Government Concerns and Anthropic’s Defense
The government’s primary concern, as interpreted by Anthropic, centered on a perceived method of bypassing, or “jailbreaking,” Fable 5. Anthropic acknowledged seeing a “specific technique” that identified “a small number of previously known, minor vulnerabilities.” The government reportedly feared these models could be exploited in ways that pose national security risks.
Anthropic, however, downplayed the severity, characterizing the vulnerabilities as “relatively simple.” They argued that other publicly available AI models could also discover these flaws without a bypass, suggesting the issue wasn’t unique. The company reiterated that “perfect jailbreak resistance is not currently possible for any model provider,” highlighting ongoing challenges in AI security.
The company stated the government’s evidence was “verbal,” detailing a potential “narrow, non-universal jailbreak.” This technique reportedly involved instructing the model to review a “specific codebase and fix any software flaws.” Anthropic maintained such capabilities are not exclusive to their new Claude models, noting similar functions in competitors like OpenAI’s GPT-5.5.
Behind the Scenes: Escalation and Official Responses
The swift ban quickly sparked extensive reporting, revealing intense activity preceding the directive. David Sacks, former White House special advisor, posted on X that Anthropic CEO Dario Amodei allegedly refused an administration request to either fix or pull the models, triggering the government’s decisive action.
Multiple outlets, including Politico, The Verge, and The Wall Street Journal, pointed to Amazon CEO Andy Jassy, or Amazon’s internal security research, as initially flagging potential vulnerabilities. Axios reported the Commerce Department gave Anthropic a mere 90 minutes to take the models offline, underscoring the perceived urgency.
Official government sentiment was also vocalized. Secretary of War Pete Hegseth, on X, validated a prior decision to expel Anthropic from Department of War premises. Chief Information Officer Kirsten Davies supported the ban, prioritizing national security over commercial interests. Conversely, an unnamed US official reportedly told The Information that these restrictions are unlikely to extend to other AI companies.
The Path Forward and a Surprising Advantage
In the wake of the directive, Anthropic dispatched senior technical staff to Washington for direct discussions with the White House. This signifies a concerted effort to address concerns and potentially restore access to their flagship models.
Anthropic expressed its belief that government should regulate unsafe deployments through transparent, fair, and technically grounded processes. However, they firmly disagreed that a “narrow potential jailbreak” justified recalling a commercial model used by millions. The company warned that such a stringent, industry-wide response could severely hinder new frontier model deployments.
Despite the immediate disruption, this incident paradoxically offers a “marketing gold” moment for Anthropic. Having the US government essentially declare your AI models too capable or too powerful for unrestricted use undeniably validates their advanced capabilities. While no company desires a ban, this situation reinforces Fable 5 and Mythos 5 as “the real deal.” Anthropic has apologized for the inconvenience, expressing hope for a “misunderstanding” and pledging to work towards restoring access.
Source: ZDNet – AI
