In an unexpected turn of events, AI powerhouse Anthropic has announced the immediate disabling of two of its cutting-edge models, Claude Fable 5 and Mythos 5. This drastic measure stems from an export control directive issued Friday afternoon by the US government, citing national security concerns. The incident marks an unprecedented moment in the evolving relationship between AI developers and government oversight.
The directive, received by Anthropic at 5:21 pm ET on Friday, initially instructed the company to suspend access for “any foreign national, whether inside or outside the United States.” To ensure absolute compliance, Anthropic, however, removed access for all its global customers. This cautious approach highlights the high stakes involved in adhering to government mandates concerning advanced AI technologies.
Unprecedented Measures and Mounting Tensions
This latest development intensifies an already strained relationship between Anthropic and the Trump administration. Earlier this year, the Department of Defense labeled Anthropic a “supply chain risk,” effectively barring government agencies and contractors from using its powerful AI technology. Anthropic had previously sought to establish clear red lines regarding military use, leading to this contentious label and subsequent lawsuits.
The recent export control directive escalates tensions dramatically, bringing national security directly into the conversation about AI model deployment. While specific concerns remain somewhat shrouded, it underscores a growing willingness to intervene directly in private sector AI releases. This incident sets a new precedent for how quickly AI capabilities can attract government scrutiny.
The Models in Question: Claude Fable 5 and Mythos 5
Central to this controversy are Claude Fable 5 and Mythos 5. Claude Fable 5, a variant of Anthropic’s powerful Mythos AI model, was publicly unveiled just this past Tuesday. It had been specifically engineered with robust safeguards, preventing it from answering questions related to sensitive domains like cybersecurity, biology, and chemistry.
Prior to its public debut, the Mythos Preview AI model underwent a limited rollout in April, a collaborative effort Anthropic stated was conducted with the US government. This earlier release aimed to empower companies with advanced cybersecurity capabilities, strengthening their defenses. It also sought to mitigate concerns that potent technology could be maliciously exploited by bad actors.
The “Jailbreak” Allegation and Anthropic’s Defense
According to Anthropic’s blog post, the government’s letter arrived without providing specific details regarding its national security concern. However, Anthropic believes the core issue revolves around a perceived “jailbreak”—a method by which Fable 5’s built-in safeguards could potentially be bypassed. This alleged vulnerability appears to be the trigger for the government’s swift action.
Anthropic has since reviewed a demonstration of this specific bypass technique. Their assessment indicates the method identified a small number of previously known, relatively minor vulnerabilities. Crucially, the company argues these particular flaws are also discoverable by other publicly available models without requiring such a bypass.
The company strongly maintains it has implemented comprehensive and robust safeguards to significantly reduce Claude Fable 5’s misuse likelihood. Furthermore, Anthropic contends the jailbreak identified by the US government is narrow in scope and not universal. They suggest it would not make an attacker meaningfully more dangerous than with access to other existing AI models.
In their defense, Anthropic clarified that the government has only provided verbal evidence of a potential, narrow, non-universal jailbreak. This evidence, as Anthropic understands, primarily involves instructing the model to analyze a specific codebase and identify software flaws. A demonstration of this specific technique was reportedly shared with the government.
A Call for Transparency and Due Process
The company’s CEO, Dario Amodei, recently emphasized his strong support for a fair, structured, and transparent government process when blocking the release of potentially unsafe AI models. This stance highlights Anthropic’s commitment to responsible AI development while advocating for clear regulatory frameworks.
However, in their official blog post addressing the directive, Anthropic explicitly stated that “this action does not adhere to those principles.” This critique underscores their belief that the government’s intervention lacked the necessary transparency and detailed justification for such a significant decision impacting a cutting-edge AI product. Spokespeople for the White House and the US Commerce Department have yet to respond to requests for comment regarding the situation.
Source: Wired – AI
