After only four days online, Fable 5 and Mythos 5 were taken offline as they were put on the US government's export control list. Anthropic complied but did not agree.

On Friday, Anthropic announced that it had received a directive from the U.S. government, forcing it to implement comprehensive access restrictions on its two most advanced AI models, Fable 5 and Mythos 5. According to Axios on June 12th, U.S. Secretary of Commerce Lute Neck sent a letter to Anthropic CEO Dario Amodei, placing Fable 5 and Mythos 5 under export control, restricting access to all institutions and individuals outside the United States, as well as foreign nationals within the country, including Anthropic's foreign employees. Anthropic stated that the directive was delivered at 5:21 PM Eastern Time and did not provide specific explanations regarding national security concerns. In a statement, the company confirmed that it had closed access to Fable 5 and Mythos 5 for all users and apologized to affected customers, while expressing belief that this was a misunderstanding and working to restore access as soon as possible. Anthropic explicitly stated that it was complying with the government's legal directive, but publicly opposed the decision, arguing that recalling a commercially deployed model that had been accessible to millions of users based on "limited prison break loopholes" lacked basis and warning that extending this standard to the entire industry would essentially stall the deployment of all cutting-edge models. Later on the same day, Amazon Web Services announced that Anthropic had requested the revocation of all users' access to Claude Fable 5 and Claude Mythos 5 within all regions. All other Anthropic models were unaffected. Dean Woodley Ball, a prominent AI policy expert in the United States, commented: "I cannot judge whether this is a legal battle against Anthropic or an extreme act of national security hawks. Either way, it is simply absurd." Gary Marcus, an honorary professor at New York University and a well-known "critic" in the global AI field, stated: "Although I have reservations about generative artificial intelligence, this is not the correct approach." According to a report by Wall Street Watch, on Tuesday, June 9th, Anthropic officially released Claude Fable 5 and simultaneously launched Claude Mythos 5. Both are based on the same underlying model architecture, with Fable 5 being the first Mythos-level model open to the public, while Mythos 5 retains more complete capabilities and is currently only open to a few trusted institutions. Export control initiated by the Department of Commerce due to prison break risk According to Axios, the direct trigger for the U.S. Department of Commerce's action was another company claiming to have successfully bypassed the security mechanisms of the Mythos model, raising government awareness of potential national security risks. The Trump administration had previously tried to persuade Anthropic to postpone the release of the latest models, but was unsuccessful, leading to the export control letter. An official mentioned that the related models need to remain locked until the national security system of the United States is reinforced, a process that may take several weeks. According to the Department of Commerce's letter, export, re-export, and domestic transfer of Anthropic models require a license, and the company must also submit an individual verification license application, with violations facing financial penalties and civil sanctions. Anthropic refutes: Prison break capability is widespread in the industry Anthropic expressed clear objections to the government's technical assessment in its statement. The company stated that it had reviewed demonstrations of related prison break techniques, identified very few and minor vulnerabilities, and noted that other publicly available models could achieve similar results without bypassing protective measures. Anthropic further pointed out that the demonstrated "narrow, non-general prison break" case provided by the government was essentially requesting the model to read specific code libraries and fix software vulnerabilities. The company verified that this capability was also prevalent in models like OpenAI's GPT-5.5, widely used by cybersecurity professionals every day, and promised to disclose more details within 24 hours. Anthropic also emphasized that in the weeks leading up to the launch of Fable 5, it had conducted thousands of hours of red team testing in collaboration with the U.S. government, the UK Artificial Intelligence Safety Institute, several private third-party organizations, and an internal team. The results showed that Fable 5's security protection was significantly better than any previously deployed model, with no testing personnel discovering a general prison break method. Compliance enforcement and policy criticism in parallel Despite raising objections, Anthropic chose to fully comply. The company stated in its declaration that it had closed access to Fable 5 and Mythos 5 for all users and apologized to affected customers, while expressing belief that this was a misunderstanding and working to restore access as soon as possible. Anthropic also issued a broader policy warning regarding regulatory frameworks, believing that if this enforcement standard were extended to the entire industry, it would effectively stall the deployment of new models by all cutting-edge model providers. The company stated that the U.S. government should have the right to prevent the deployment of insecure models, but the procedures should be transparent, fair, clear, and based on technical facts, and "this action does not meet these principles." Further deterioration of relations between Anthropic and the government This incident further strained the already complex relationship between Anthropic and the Trump administration. As Anthropic prepared for an IPO, there had been signs of easing tensions between the two parties over the past few months. However, the current situation indicates a renewed deterioration in government-business relations. Earlier this month, President Trump signed an executive order requiring major AI developers to voluntarily submit their models to government cybersecurity testing before releasing the most powerful models. Anthropic had established a collaboration with the AI Standards and Innovation Center under the U.S. Department of Commerce for pre-launch testing. However, the export control measures taken this time took a completely different approach. The executive order explicitly avoided the licensing system, and White House Chief AI Advisor David Sacks had previously advocated avoiding "regulatory capture" of large labs. A U.S. government official stated that Trump "does not want to harm this industry and hopes that innovation continues." In March of this year, the U.S. Department of Defense had identified Anthropic as a "supply chain risk" company, and now the Department of Commerce has placed it under the export control license system. This article was originally published by Wall Street Watch and edited by GMTEight editor, Chen Yufeng.