Summary
- The U.S. government has mandated that Anthropic cease access to its advanced AI models, Fable 5 and Mythos 5, citing national security risks related to a potential jailbreak vulnerability.
- While Anthropic has complied with the order, it contests the government's findings, asserting that the identified vulnerability is straightforward and can be replicated with other publicly available models like GPT-5.5.
- The firm cautioned that the directive could create a troubling precedent, potentially stalling the deployment of new AI models across the industry.
On Friday, the U.S. government implemented an emergency export control order, instructing Anthropic to immediately restrict access to its two leading artificial intelligence models—Claude Fable 5 and Claude Mythos 5—both launched just days prior for all foreign nationals, including Anthropic's own staff, due to national security issues.
This directive prohibits any foreign individual from accessing the models, regardless of their location, which compelled Anthropic to deactivate the models for its entire clientele to ensure compliance.
The communication from the government did not detail specific national security concerns but suggested that officials had become aware of a method to bypass the publicly available Fable 5 model. The Mythos 5 model, known for its fewer restrictions and strong capabilities in cybersecurity exploitation, was limited to select partners.
Anthropic questioned the seriousness of the government's findings, stating that after reviewing a demonstration of the alleged technique, they found the vulnerabilities to be relatively simple and that other publicly available models can identify them without needing any bypass.
The company noted that the government has only provided verbal evidence of a potential narrow, non-universal jailbreak, which essentially involves asking the model to analyze a specific codebase and remedy any software defects. Anthropic confirmed that the capabilities demonstrated are already commonly found in competing models, including OpenAI's GPT-5.5.
Although Anthropic is complying with the order, it expressed concern that this action establishes a dangerous precedent. "If this standard were to be adopted throughout the industry," the company stated, "it would effectively stop all new model launches for all frontier model developers."
Access to Anthropic's other models remains unaffected, and the company is actively working to restore access as quickly as possible.
I’ve had several discussions with individuals both inside and outside the government regarding the current situation with Anthropic, and here's my perspective:
— As we know, Anthropic publicly released its Mythos class models earlier this week under the commercial name Fable.…
— David Sacks (@DavidSacks) June 13, 2026
On Saturday, David Sacks, co-chair of the President’s Council of Advisers on Science and Technology, stated on X that “a highly credible, trusted partner of both Anthropic and the U.S. government who was testing Fable came forward with a jailbreak of those guardrails. The administration requested [Anthropic CEO Dario Amodei] to rectify the jailbreak or retract the model. Dario declined.”
Sacks also claimed that Anthropic’s response to the government's request contradicted the company's public statements regarding the necessity for AI safety and regulation—a viewpoint that Amodei reiterated earlier this week in a blog post.
“Anthropic prioritized the ongoing availability of the consumer model over safety,” Sacks remarked. “In response, the administration issued the export control. They did so with reluctance and were surprised by Anthropic’s lack of cooperation regarding a reasonable safety request (i.e., addressing the jailbreak issue). Anthropic’s response is significantly at odds with their branding and ethos as a safe AI research entity.”
Sacks expressed hope that Anthropic would resolve the issue and that Fable 5 could return to public availability.
“The administration values Anthropic's technical expertise and believes this issue, while serious, should be easily rectified,” he stated. “The ball is in Anthropic’s court.”
Earlier this year, tensions arose between Anthropic and the U.S. government when the AI firm declined to sign an expanded agreement that would permit mass domestic surveillance of Americans and fully autonomous lethal weapons systems.
President Donald Trump criticized the company after negotiations collapsed, and the Department of Defense labeled Anthropic a “supply chain risk”—a classification that the firm has challenged in court. Since then, reports indicate that the tension between the two parties has eased as the government expressed interest in utilizing Claude Mythos and other models.