Tldr:
- The United States Government has issued an Export Control Directive ordering Anthropic to suspend all access to Fable 5 and Mythos 5 globally.
- Anthropic reviewed the jailbreak report and found that the capabilities were already available in models like OpenAI’s GPT-5.5.
- The reported jailbreak involved requiring Fable 5 to read the code base and report software flaws, without revealing any malicious result.
- Anthropic warned that the recall standard, if implemented industry-wide, would effectively halt all deployments of frontier AI models.
Anthropic Disabled global access to Fable 5 and Mythos 5 following US government export control directives.
The order, which was received at 5:21 PM ET, cites national security concerns related to the reported jailbreak method. All other human models remain available.
The company says it adheres to the directive while contesting the technical basis of the decision.
Targets of government guidance on how to escape from prison have been reported
US government Issued Directive without disclosing specific national security details in writing. Officials reported verbally that they knew of a method capable of bypassing the safeguards of Myth 5.
Anthropic reviewed a demo of this technology and found that it revealed only minor previously known vulnerabilities.
The company reviewed what it believes is the report behind the government’s decision. Anthropic stated that the level of ability offered “Available across a range of other models, including OpenAI’s GPT-5.5, and used daily by defenders who keep systems secure.” That review found no Fable-specific increase in results.
Reported jailbreaking basically involves asking the model to read the code base and identify software flaws. This was confirmed by Anthropy “He did not even receive detection of a potential non-global jailbreak that resulted in a malicious result.” The potential jailbreaks uncovered were either completely benign or were classified as minor findings.
The directive requires the suspension of access of all foreign nationals, including humanitarian personnel with foreign national status, inside and outside the country. US. The company said compliance means disabling the forms for all customers to avoid any breach of the order.
Anthropic disputes the standard applied to commercial models
Anthropy took off Myth 5 With a defense-in-depth strategy, it combines tight jailbreak resistance, real-time monitoring, and mandatory 30-day data retention.
“Perfect jailbreak resistance is not currently possible for any model provider,” the company admitted during the launch.
The 30-day data retention policy was a deliberate trade-off. This was met with pushback from customers but allowed Anthropic to quickly detect, study, and respond to jailbreak attempts.
Anthropic described this as making jailbreaks “either cramped or very expensive to produce,” making risk levels comparable to other models deployed throughout the industry.
Regarding the government’s authority to act, Anthropic said:It believes the government should have the ability to prevent unsafe deployments, as part of a legal process that is transparent, fair, clear, and grounded in technical facts. The company said this guidance does not meet these standards.
Anthropic He cautioned against applying this recall standard too broadly “It would essentially halt all new model deployments for all frontier model providers.”
The company committed to releasing additional technical details within 24 hours and confirmed that all other models in its range continue to operate without restrictions.
