Unprecedented! US Government Forcibly "Recalls" Commercial Model Due to Jailbreak Risk, Anthropic Overnight Takes Down Fable 5 and Publicly Protests

According to Scryer Beating monitoring, the U.S. government, citing national security authority, has issued an export control directive to the AI security company Anthropic, suspending all foreign nationals (whether located within or outside the U.S., including Anthropic's foreign employees) from accessing its latest Fable 5 and Mythos 5 models. To ensure legal compliance, Anthropic announced that it was compelled to immediately and unexpectedly disable the aforementioned two models for all global customers. Access to other Claude series models (such as 3.5 Sonnet, Haiku, etc.) remains unaffected.

The directive was delivered at 5:21 p.m. Eastern Time today (June 12), with no specific details provided on the national security risks. Anthropic disclosed that the government made this decision upon learning of a potential "jailbreak" method that could bypass the security protections of Fable 5. However, Anthropic stated after evaluation that this method could only be used to explore a few known minor vulnerabilities (such as reading specific code libraries and fixing software bugs). Such capabilities are widespread in the industry, and other publicly available models (such as OpenAI's GPT-5.5) can achieve this without jailbreaking, which is also a common practice for cybersecurity professionals.

Prior to the release of Fable 5, Anthropic engaged in cumulative red team testing with the U.S. government, the U.K. AI Security Institute (AISI), and several third-party organizations for thousands of hours. The tests indicated that Fable 5's defense system is stronger than previous models, and as of now, no widespread "universal jailbreak" that can bypass defenses has been discovered. Anthropic advocates for a "defense-in-depth" strategy and requires in its terms of service the retention of user data for 30 days to detect and prevent jailbreaks, although this policy increases compliance costs for customers.

Anthropic emphasized that while the company is complying with the legal directive to shut down the models, it strongly opposes recalling commercially deployed models to hundreds of millions of users based on a single narrow potential jailbreak. Applying this standard universally would stall all new deployments from leading-edge model providers. Anthropic urged that government interventions in model deployment should follow a transparent, fair, clear, and tech-fact-based due process, which the current directive does not align with. The company deeply apologized for the service interruption caused by this and considered it a "misunderstanding," actively engaging with the government to seek a prompt restoration of access.