The U.S. government has compelled Anthropic to immediately disable access worldwide to its two most advanced AI models, Claude Fable 5 and Claude Mythos 5, citing national security concerns tied to potential cybersecurity vulnerabilities. Anthropic has expressed strong disagreement with the decision, emphasizing its extensive safety protocols and arguing the move could stifle innovation in frontier AI development.
- US government shuts down Anthropic’s Claude Fable 5 and Mythos 5 citing security fears
- Anthropic disputes government claims, emphasizing extensive safety and narrow risk
- Decision affects global users, raising industry-wide regulatory and innovation questions
What happened
On Friday, the U.S. government issued a directive requiring Anthropic to disable access to two of its most advanced AI models: Claude Fable 5 and Claude Mythos 5. The order, framed as an export control action, was aimed at restricting foreign national access but resulted in a worldwide shutdown. Anthropic complied promptly but expressed disagreement with the move in public statements. The models targeted include Mythos, Anthropic’s highly capable cybersecurity-focused AI shared with select partners under Project Glasswing, and Fable 5, a safer, commercially available variant.
The government based its decision on concerns about a potential, narrowly defined jailbreak vulnerability in Fable 5, one that purportedly allows the model to identify software flaws after being prompted in a specific way. Anthropic argues the underlying capability is already common in other major AI systems and that they have implemented comprehensive, independent safety classifiers to guard against harmful outputs. Despite these assurances, access to these models was cut off to all users globally.
Why it matters
This government action underscores the tension between pioneering AI development and national security worries. Anthropic positioned Mythos as dangerously powerful, restricting it with a controlled program due to its ability to expose software vulnerabilities across major systems. By contrast, Fable 5 was introduced as a model safe enough for broad use, integrating strong guardrails to mitigate sensitive risks. The shutdown disrupts access to arguably the most capable AI model available to the public and complicates Anthropic’s commercial and public safety reputation.
The decision highlights wider regulatory challenges facing AI providers, as it targets a narrow security issue but has sweeping effects. Anthropic warned that if such restrictive standards were consistently applied, new AI model deployments could stall across the industry. The incident also reveals the delicate balance between transparency about AI risks and inviting government scrutiny that could hamper innovation and market competition, especially as Anthropic is expected to pursue an IPO.
What to watch next
Industry observers will be closely monitoring how Anthropic responds legally and strategically to the shutdown directive and whether it can negotiate terms to restore model access. The company’s arguments about the narrow scope of the alleged jailbreak and its safety system design will be key points in any regulatory discussions. Other AI providers, including major competitors like OpenAI, may also adjust their messaging and risk management in light of this precedent.
More broadly, the U.S. government’s approach to export control and AI security regulation is poised to be a defining factor in shaping future AI innovation and deployment in the US and internationally. Stakeholders will look for clarifications on standards, transparency requirements, and how concerns about misuse or security risks can be balanced with commercial availability to prevent disruptions for users and developers.