Anthropic's Safety Warnings Backfire as US Government Halts Access to Its Most Powerful AI Models

The U.S. government has delivered a significant blow to AI developer Anthropic, ordering the immediate shutdown of access to two of its most powerful artificial intelligence models, Claude Fable 5 and Claude Mythos 5. Citing national security concerns, the directive, issued on a Friday evening, mandates that Anthropic disable both models for all users worldwide. While Anthropic has publicly confirmed its compliance, the company has made it clear that it fundamentally disagrees with the government's assessment and the rationale behind the unprecedented move. The government's order, framed as an export control action primarily aimed at restricting foreign national access, has far broader implications, forcing a global cessation of services for these models. The core of the controversy revolves around Mythos 5, which Anthropic had previously showcased as its most capable AI model. Previewed in early April, Mythos 5 was kept under tight restrictions due to its exceptional ability to identify security vulnerabilities across major operating systems and web browsers. Instead of a broad release, Anthropic launched Project Glasswing, a controlled program where Mythos 5 was shared with approximately 50 vetted organizations, including tech giants like Amazon, Apple, Google, Microsoft, and cybersecurity firm CrowdStrike, specifically for defensive cybersecurity applications. Just three days prior to the government's intervention, Anthropic had released Fable 5, a commercially viable version of Mythos fitted with crucial guardrails. These safeguards were designed to block responses in high-risk areas such as cybersecurity and biology, making it suitable for general public release. According to benchmark tests from Vals AI, a company specializing in tracking AI performance, Fable 5 immediately became the most capable AI model available to the public. However, Anthropic states that its understanding is that the underlying concern for the government's action is a claimed jailbreak of Fable 5, for which only verbal evidence of a "potential narrow, non-universal jailbreak" has been provided. Anthropic vehemently defends its models, arguing that the alleged jailbreak involves prompting the model to read specific codebases and identify software flaws – a level of capability, it asserts, that is already widely available in other publicly accessible models, including OpenAI's GPT-5.5. Furthermore, such capabilities are routinely utilized by cybersecurity professionals for legitimate defensive purposes. The company emphasizes that its strongest safeguards operate through independent classifier systems, separate from the model itself, ensuring that even if a user bypasses a refusal, the core protections against dangerous outputs remain intact. Anthropic expressed deep 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," and warned that applying such a standard across the industry would "essentially halt all new model deployments for all frontier model providers." This incident carries a profound irony for Anthropic, a company widely expected to pursue an IPO this year and which has meticulously cultivated an identity as the safety-conscious alternative in the competitive AI landscape. Its very caution in restricting Mythos, promoting it as a model too dangerous for public release, appears to have inadvertently drawn the intense government scrutiny that now threatens to disrupt its business. Observers recall Sam Altman of OpenAI's previous comments, who in April described Anthropic's handling of Mythos as "fear-based marketing," likening it to selling a bomb shelter after building a bomb. While Altman didn't predict a government shutdown, his observation that publicizing an AI as uniquely dangerous tends to make the world – including governments – listen, has certainly come back to bite Anthropic.