Export control directive forces immediate shutdown of Fable 5 and Mythos 5 models based on claimed national security concerns around code vulnerability detection.
Summary
- US government issued export control directive forcing Anthropic to suspend access to Fable 5 and Mythos 5 models for all foreign nationals
- Government cites national security concerns over alleged jailbreaking technique that identifies software vulnerabilities
- Anthropic says the demonstrated capability is widely available in other public models including OpenAI’s GPT-5.5
- Access was cut off at 9:59pm ET on June 12, 2026, with no advance warning to customers
The US government has ordered Anthropic to immediately suspend access to its Fable 5 and Mythos 5 AI models, citing national security authorities under export control regulations. The directive, received at 5:21pm ET, requires blocking access for all foreign nationals, including Anthropic’s own international employees.
Limited Technical Details
According to Anthropic, the government’s concern centres on a purported jailbreaking method that allows the models to identify software vulnerabilities. The company states it reviewed a demonstration of this technique being used to find “a small number of previously known, minor vulnerabilities” that other publicly available models can also detect without requiring any bypass.
The alleged jailbreak technique reportedly involves asking the model to read specific code and identify software flaws. Anthropic notes that this capability is “widely available from other models (including OpenAI’s GPT-5.5), and is used every day by the defenders who keep systems safe.”
Abrupt Implementation
The shutdown was implemented without advance notice to customers. Anthropic confirmed that access to claude-fable-5 was terminated at 9:59pm ET, approximately four hours after receiving the government directive. All other Anthropic models remain available.
The government directive provided no specific details about its national security concerns beyond the alleged jailbreaking capability. Anthropic states it has only received “verbal evidence of a potential narrow, non-universal jailbreak” and promises to share additional details within 24 hours.
Why it matters
This unprecedented government intervention demonstrates how AI capabilities that seem routine for security professionals could trigger export controls without warning. CISOs using advanced AI models for vulnerability assessment or code review need contingency plans for sudden service disruptions, particularly when working with international teams or on systems that might be classified as dual-use technology.
What to do now
- Review your organisation’s dependence on specific AI models for security operations and vulnerability management
- Develop alternative approaches for code analysis and vulnerability detection that don’t rely on single AI providers
- Monitor Anthropic’s promised updates over the next 24 hours for additional technical details
