US Government Forces Anthropic to Shut Off Flagship AI Models Globally
The US Department of Commerce has ordered Anthropic to immediately block all foreign nationals from accessing Fable 5 and Mythos 5, the company's most capable frontier AI models. Faced with the order, Anthropic chose to suspend both models worldwide rather than attempt to enforce nationality restrictions on its global user base.
The action marks one of the most aggressive applications of AI export control authority to date, drawing sharp condemnation from AI researchers, civil liberties advocates, and international partners.
What Happened
According to reporting by BleepingComputer, the US government cited a discovered jailbreak technique capable of eliciting sensitive cybersecurity assistance from both Fable 5 and Mythos 5. Commerce Department officials characterized the models as posing unacceptable national security risks under existing export control frameworks if accessible to foreign nationals.
Anthropic confirmed compliance with the order but publicly challenged the government's rationale. The company stated the cited jailbreak is "narrow" in scope and that comparable capabilities are "widely available elsewhere" — including in open-weight models from international competitors.
The company suspended Fable 5 and Mythos 5 globally rather than implement nationality-based access controls, which Anthropic leadership indicated would be technically unworkable and legally complex.
Industry Reaction
The move generated immediate backlash from the AI research community:
- Security researchers warned that restricting access to Anthropic's models primarily disadvantages legitimate defensive research while having minimal impact on adversarial actors who can access equivalent capabilities elsewhere
- International partners expressed concern that the action signals a broader trend toward AI nationalism that could fragment global research collaboration
- Industry analysts questioned whether the jailbreak cited constitutes genuine unique risk given the proliferation of capable open-source models
Researchers noted the irony that suspending Fable 5 and Mythos 5 — which include robust safety guardrails — may push foreign researchers toward less safety-conscious alternatives.
Background: AI Export Controls
The Commerce Department has been expanding its interpretation of existing export control regulations to cover frontier AI models, arguing that certain AI capabilities constitute "dual-use technology" subject to Export Administration Regulations (EAR).
Previous AI-related export restrictions have focused on semiconductor hardware (particularly GPU exports to China) and specific software tools. Applying export controls to AI model access via API represents a significant expansion of the regulatory approach.
Anthropic's Position
Anthropic has framed its compliance as temporary and under protest. The company indicated it is:
- Cooperating fully with the Commerce Department order
- Engaging with government officials to present evidence on the narrow scope of the cited vulnerability
- Advocating for a risk-proportionate approach that distinguishes between legitimate research use and adversarial exploitation
The company did not announce a timeline for restoring access to Fable 5 or Mythos 5 for international users.
What This Means for Users
- US-based users: Access to Fable 5 and Mythos 5 through Anthropic's API and Claude.ai is unaffected
- International users: Both models are suspended globally pending resolution of the export control dispute
- Enterprise customers: Organizations with Anthropic enterprise agreements outside the US are effectively cut off from the company's most capable models
- Researchers: Academic and security researchers outside the US face disruption to projects relying on frontier model access
Broader Implications
The episode raises fundamental questions about how AI export controls will be structured as frontier models become increasingly central to critical industries:
- Verification challenges: Nationality-based access controls for web APIs have no reliable technical enforcement mechanism
- Market fragmentation: Broad restrictions may accelerate the development of competing frontier models by governments and companies outside US jurisdiction
- Security research impact: Defensive cybersecurity research increasingly depends on access to frontier AI capabilities; blanket restrictions undermine this work
The incident is expected to intensify congressional and regulatory debate over the appropriate scope of AI export controls heading into the second half of 2026.