Anthropic Halts Fable 5 and Mythos 5 AI Models After U.S. Government Warning
Anthropic ceased global access to its Claude Fable 5 and Claude Mythos 5 AI models following an enforcement order from the U.S. Commerce Department. The order, issued on June 12, restricted access by foreign nationals, citing national security concerns related to potential misuse for cybersecurity vulnerabilities. Anthropic opted for a full shutdown, stating it could not immediately differentiate between eligible and ineligible users. The decision came after Amazon, Anthropic's largest investor, along with other tech companies, raised concerns with the Trump administration on June 11 regarding the models' susceptibility to 'jailbreaking' for risky cybersecurity queries. Despite Anthropic's existing safeguards and initial reluctance to suspend the models due to a perceived lack of detailed technical evidence, the Commerce Department issued an ultimatum before imposing the formal restrictions.
Anthropic has shut down global access to its new Claude Fable 5 and Claude Mythos 5 AI models. The action followed an enforcement order issued by the U.S. Commerce Department on Friday, June 12, which restricted use of the models by foreign nationals, citing national security concerns.
The company stated that it disabled access globally because the order made it impossible to immediately distinguish between eligible and ineligible users. This extensive measure was taken out of caution and a commitment to compliance.
The chain of events began on Thursday, June 11, when Amazon, a major investor in Anthropic, contacted senior White House and administration officials. Amazon researchers reportedly identified a method to 'jailbreak' the Fable 5 and Mythos 5 models, potentially coercing them into assisting with risky cybersecurity questions. Similar concerns were reportedly raised by at least five other technology companies around the same time.
White House and Commerce Department officials responded to these concerns with urgency, fearing that foreign entities could bypass Anthropic's security measures to identify and exploit software vulnerabilities. Administration officials reportedly spent hours on Friday morning, June 12, attempting to persuade Anthropic to voluntarily suspend access to the models.
Anthropic initially declined to voluntarily pause the models' release. The company stated publicly that the government did not provide detailed technical evidence of the alleged issue, with concerns involving only verbal descriptions. In a blog post on June 12, Anthropic noted, "We reviewed a demonstration of this specific technique being used to identify a small number of previously known, minor vulnerabilities. These vulnerabilities all appear relatively simple, and we have found that other publicly available models are able to discover them as well without requiring a bypass."
By approximately 1 p.m. ET on June 12, Anthropic received a call from administration officials, warning that export restrictions would be imposed unless the company took action. Anthropic was reportedly given about 90 minutes to comply. The Commerce Department's enforcement order arrived at 5:21 p.m. ET on Friday.
The Fable 5 and Mythos 5 models are the first public releases derived from Anthropic's Mythos family. During training, the original Mythos model demonstrated a capacity for finding and exploiting software bugs. This led Anthropic to implement safeguards, routing prompts deemed sensitive in areas like cybersecurity, biology, and chemistry to a less capable model, Claude Opus 4.8, which has its own guardrails.
According to Fast Company, Anthropic's internal and external 'red teams' had spent over 1,000 hours attempting to bypass the Fable 5's filters, reporting no success in identifying universal jailbreaks. However, at least one independent AI researcher reportedly claimed to have found a successful jailbreak prior to Amazon's report to the government.
