On June 12, a sharp regulatory confrontation erupted between the U.S. government and Anthropic regarding the deployment of advanced artificial intelligence models. The Commerce Department issued an export control directive citing national security, mandating an immediate suspension of access to the Fable 5 and Mythos 5 models for all foreign entities. This order extends beyond international borders, encompassing foreign citizens residing within the United States and even foreign employees working directly for Anthropic. To ensure strict compliance with this broad mandate, the company was forced to completely sever access to both models for its entire user base. Data compiled by Woofun AI indicates that technical limitations prevent the distinction between foreign and domestic users within the current infrastructure, making a total shutdown the only viable compliance path. Consequently, all traffic previously routed to these models has been automatically redirected to the Claude Opus 4.8 version, while other model offerings remain operational.

The abrupt cessation of service came just three days after the official public release of Fable 5 and Mythos 5 on June 9, sending shockwaves through the technology and AI sectors. Mythos represents a new model family from Anthropic that surpasses the capabilities of the Opus series, marking the highest tier of publicly deployable intelligence currently available. The initial iteration, Claude Mythos Preview, launched in April under the 'Project Glasswing' initiative with access restricted to select partners due to its potent cybersecurity applications. Fable 5 stands as the first Mythos-level model opened to the general public, achieving top-tier performance across software engineering, knowledge work, visual understanding, and scientific research benchmarks. To mitigate risks, Anthropic integrated dedicated security mechanisms into Fable 5 that automatically block responses in high-risk domains such as biology and chemistry, reverting queries to Claude Opus 4.8 for processing.

Mythos 5 operates on the same underlying architecture but features reduced security protections, intended exclusively for previously approved institutions involved in Project Glasswing. Positioned as a professional tool for cybersecurity defenders and critical infrastructure operators, both models carry identical pricing structures at $10 per million input tokens and $50 per million output tokens. The trigger for the government's intervention appears to stem from a claim by a third party that it could 'jailbreak' the Mythos model, raising alarms within the Trump administration regarding potential national security vulnerabilities. Reports suggest the administration had previously attempted to block the release of these models unsuccessfully, prompting the shift to the more stringent measure of export controls. Woofun AI notes that the directive was formalized via a letter from Commerce Secretary Howard Lutnick to Anthropic CEO Dario Amodei on June 12.

In response, Anthropic issued a rare, strongly worded statement that systematically refuted the government's rationale while affirming its commitment to compliance. The company argued that the 'jailbreak' evidence cited by the government pertains to a narrow, non-generalizable attack method requiring the model to read specific code libraries and fix software vulnerabilities. Anthropic emphasized that this capability exists in other publicly available models, including OpenAI's GPT-5.5, and is routinely utilized by cybersecurity defenders for standard system maintenance. The firm warned that if the existence of a limited potential jailbreak becomes the standard for recalling deployed commercial models, it would effectively halt the new deployment of all cutting-edge models across the industry.

Furthermore, the company criticized the government's action for bypassing transparent, fair, and fact-based legal procedures previously advocated for in public discourse.

The full statement released by Anthropic detailed the timeline and technical context of the incident, noting receipt of the directive at 5:21 PM Eastern Time. The company reviewed the demonstration of the alleged bypass technique and found it capable of identifying only a small number of previously known, minor vulnerabilities that are relatively simple to detect. Woofun AI analysis suggests that these findings do not reflect unique risks inherent to the Mythos architecture. Anthropic highlighted that in the weeks leading up to the Fable 5 release, it collaborated with the U.S. government, the UK AI Safety Institute, and third-party organizations to conduct thousands of hours of red team testing. These tests confirmed that Fable's security protections significantly outperform any previously deployed models, with no testers able to discover a 'universal jailbreak' method capable of comprehensively breaching security protections.

Acknowledging that perfect jailbreak protection is currently unfeasible across the industry, Anthropic outlined its layered defense strategy designed to make attacks either narrow in scope or prohibitively costly. This approach is supplemented by comprehensive monitoring and a policy requiring customer data retention for 30 days to facilitate the study and containment of jailbreak risks. The company stated that it has not received formal disclosures regarding non-general jailbreaks leading to harmful consequences, characterizing the potential cases disclosed as either harmless or minor. While complying with the directive to shut down access, Anthropic expressed disagreement with the premise that a limited potential jailbreak justifies recalling a model deployed to hundreds of millions of users. The firm apologized for the inconvenience and indicated it would release further technical details within the next 24 hours while actively working to restore access as soon as possible.