TLDR:
- US export control order forced global shutdown of Fable 5 and Mythos 5 models immediately across all users
- Anthropic says restriction stems from alleged jailbreak concern but calls issue narrow and non-systemic in scope
- Other Claude models remain online as company complies with government national security directive requirements
- Firm disputes severity of claims, citing prior red-teaming tests and absence of verified harmful exploits
A US government export control directive has forced Anthropic to suspend global access to Fable 5 and Mythos 5. The order applies to all foreign nationals, including employees outside the United States.
The decision triggered an immediate shutdown of both models across all customer environments. According to internal communication, the directive was issued citing national security concerns tied to potential jailbreak vulnerabilities.
Fable 5 and Mythos 5 Export Control Order Shakes Anthropic AI Access
AnthropicAI confirmed it received the directive at 5:21pm ET from US authorities.
The order required immediate suspension of Fable 5 and Mythos 5 access worldwide. The restriction applies even to foreign national staff working within the company.
The company stated compliance was mandatory under export control rules.
As a result, it disabled both models across its infrastructure. Other Claude models remain fully operational without restriction.
The announcement highlighted that the directive affects users across all regions. Customers outside the United States lost access at the same time as domestic users. The scope of the order reflects broad national security classification.
Anthropic noted the shutdown was abrupt and not pre-planned.
Engineering teams executed global deactivation procedures shortly after receiving the notice. Service disruption affected enterprise users and developers relying on the models.
Fable 5 and Mythos 5 Security Concerns and Jailbreak Claims Explained
The directive reportedly stemmed from concerns about a potential jailbreak method targeting Fable 5.
Authorities believed the technique could expose cybersecurity-related capabilities under certain conditions. The company reviewed the same demonstration internally.
Anthropic stated the identified issue involved narrow vulnerabilities already seen in other models. It added that similar weaknesses could be reproduced using publicly available AI systems. The company did not identify evidence of a universal jailbreak.
Internal assessments showed safeguards were tested extensively before release.
The models underwent thousands of hours of red-teaming with external partners. These included government-linked AI safety institutes and third-party evaluators.
According to Anthropic, no verified harmful deployment resulted from the reported vulnerability. The company said it had not received documentation of a broad exploit affecting model safety systems. It described the issue as limited in scope and non-systemic.
Despite disagreement with the directive’s severity, Anthropic complied with legal requirements.
The firm emphasized ongoing discussions with regulators to restore access. It also reaffirmed that other models remain unaffected and continue operating normally.
