In brief
- Washington has directed Anthropic to cut off access to its two newest AI systems, Fable 5 and Mythos 5, pointing to a possible jailbreak flaw that poses a national security risk.
- Anthropic followed orders but challenged the conclusion, saying the flaw is straightforward and can already be replicated with other widely available models like GPT-5.5.
- The firm cautioned that this order could set a troubling standard that, if rolled out across the industry, may freeze the launch of any new cutting-edge AI models.
On Friday, the U.S. government issued an urgent export control order directing Anthropic to immediately shut off access to its two most advanced AI models—Claude Fable 5 and Claude Mythos 5, both of which had been made public just days before—for every foreign national, including the company’s own overseas staff, on national security grounds.
The order prohibits any person who is not a U.S. citizen from interacting with these models, regardless of their location. Because of how broad the directive is, Anthropic had to turn off the models for its whole range of customers to remain in compliance.
The government’s letter did not spell out the exact nature of the national security concern, but officials reportedly learned of a way to circumvent—or “jailbreak”—the openly accessible Fable 5 model. Mythos 5, which operates with fewer safety restrictions and excels at uncovering cybersecurity vulnerabilities, had only been shared with a limited set of partners.
Anthropic pushed back on how serious the finding actually is. After examining a demonstration of the technique, the company determined that the weaknesses involved are fairly basic and that other widely available AI models are equally capable of exploiting them without any need for additional bypass methods.
According to Anthropic, the government has so far offered only verbal accounts of what amounts to a narrow, non-universal jailbreak—essentially involving instructing the model to review a particular codebase and patch any software bugs. The company confirmed that the capabilities on display are already readily accessible through competing models, such as OpenAI’s GPT-5.5.
Although Anthropic is complying with the directive, the company voiced concerns about the broader implications. “Were this standard to be enforced across the tech sector,” the organization stated, “it would effectively bring a standstill to all new model releases from every leading AI developer.”
All other Anthropic models remain unaffected by the order. The company pledged to restore full access at the earliest opportunity.
I’ve spoken with a number of individuals both within and outside the government regarding the current situation with Anthropic, and here’s what I consider to be accurate:
— As everyone is aware, Anthropic publicly launched its Mythos-class models earlier this week under the commercial brand Fable.…
— David Sacks (@DavidSacks) June 13, 2026
On Saturday, David Sacks—who serves as co-chair of the President’s Council of Advisers on Science and Technology—posted on X that “a highly reliable, trusted partner to both Anthropic and the U.S. government, who was evaluating Fable, flagged a way to circumvent those guardrails. The administration asked [Anthropic CEO Dario Amodei] to fix the flaw or pull the model from circulation. Dario declined.”
Sacks went on to argue that Anthropic’s response flew in the face of the company’s own public messaging about the importance of AI safety and regulation—a position that Amodei had once again reiterated this week in a blog post.
“Anthropic chose to keep offering the consumer product rather than address safety concerns,” Sacks wrote. “In response, the administration issued the export control. It did so reluctantly. The government has been genuinely surprised that Anthropic has been unwilling to work with a reasonable safety request—namely, fixing the jailbreak issue. Anthropic’s behavior stands in stark contrast to its public image and reputation as a safety-focused AI research organization.”
Sacks also mentioned that the administration hopes Anthropic will resolve the issue and that Fable 5 can eventually return to public availability.
“The administrator recognizes Anthropic’s engineering expertise and believes that this matter, though significant, should be straightforward to fix,” he stated. “It’s now up to Anthropic to take action.”
Earlier this year, Anthropic and the U.S. government clashed when the AI company declined to enter into an expanded contract permitting widespread domestic surveillance of American citizens and fully autonomous lethal weapons systems.
President Donald Trump lashed out at the company following the collapse of negotiations, and the Department of Defense designated Anthropic as a “supply chain risk”—a classification that the company has contested through legal channels. In the time since, reports have indicated that tensions between both parties have eased as the government pursued access to Claude Mythos and other models.
Daily Debrief Newsletter
Kick off each morning with the most important news happening now, along with exclusive features, a podcast, videos and much more.
