The US Commerce Department delivered an export control directive to Anthropic on June 12, 2026, ordering the company to immediately restrict access to its two most powerful AI models — Claude Fable 5 and Claude Mythos 5 — from all foreign nationals, regardless of where they are located.
Because Anthropic cannot verify users' nationalities in real time, the company took the only compliant path available: a complete global shutdown of both models for all customers.
What Happened
Commerce Secretary Howard Lutnick sent a letter to Anthropic CEO Dario Amodei on June 1, 2026, laying out the export control framework. The directive was formally delivered to the company at 5:21 PM ET on June 12 — just three days after Anthropic launched Fable 5 and Mythos 5 to the public.
In its official statement, Anthropic confirmed the scope of the action:
"The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance."
Three systems are affected: Claude Mythos 5, Mythos Preview, and the Fable 5 variant. Every other Anthropic model — including Claude Opus 4.8, Sonnet 4.6, and Haiku 4.5 — remains fully operational.
The Trigger: A Claimed Jailbreak
The Commerce Department cited national security concerns rooted in the models' advanced cybersecurity capabilities. The trigger was a claim by an unnamed rival company that it had discovered a method to "jailbreak" Mythos 5 — bypassing its safety guardrails in ways the administration deemed dangerous.
According to reporting by Axios, the Trump administration had previously attempted to block the public release of Mythos 5 but failed. The export control directive followed as the next available lever.
Prior to the suspension, Mythos 5 had been distributed through "Project Glasswing," a controlled program limiting access to select US firms — including Microsoft, Google, and NVIDIA — for addressing cybersecurity vulnerabilities.
Anthropic Disputes the Rationale
Anthropic is complying with the directive while publicly contesting its basis. The company says the government has provided only "verbal evidence of a narrow, non-universal jailbreak" that involves asking the model to read a specific codebase and identify software flaws.
Anthropic argues this capability is already available in other frontier models, specifically citing OpenAI's GPT-5.5, and that applying this standard broadly would "effectively halt new model deployments from every frontier provider." The company called the situation "a likely misunderstanding" and said it is actively working to restore access.
Developer and Business Impact
For developers and businesses that integrated Fable 5 or Mythos 5 into their workflows, the impact is immediate:
- Existing active sessions are terminating with errors
- New sessions automatically default to Claude Opus 4.8 or other available models
- API requests to Fable 5 return errors, requiring integrations to be updated
The disruption is particularly acute for enterprise customers who adopted the Mythos tier for high-stakes applications in code analysis, legal research, and cybersecurity.
A Landmark Precedent
Legal and policy experts are watching this case closely. This marks the first time US export controls have targeted deployed commercial AI software — not chips, not hardware, but a running cloud service. The precedent carries enormous implications:
- Frontier AI models may increasingly be treated as strategic assets equivalent to semiconductors or advanced military systems
- International development teams with non-US members could face systematic restrictions
- Global enterprises relying on US-based AI providers may need sovereign or open-source alternatives
The order specifically prohibits "any export, re-export, or domestic transfer of these models to non-US persons" — meaning even Anthropic's own foreign-national employees cannot access Fable 5 or Mythos 5.
What This Means for MENA Businesses
For businesses in the MENA region that have adopted Anthropic's top-tier models, the shutdown is an immediate operational disruption. It also surfaces a structural vulnerability in relying exclusively on US-controlled AI services for mission-critical workflows.
The episode reinforces the case for diversified AI stacks — combining open-weight models with sovereign infrastructure — particularly in regions subject to US export jurisdiction. Providers like Cohere, Mistral, and locally deployable open-source models now become more strategically attractive as fallback options.
Anthropic says it is working with the Commerce Department to resolve what it characterizes as a misunderstanding. Until then, Fable 5 and Mythos 5 remain offline globally.
Source: Axios