Anthropic has publicly stated that it took Fable 5 and Mythos 5 — its two most recently released frontier AI models — offline specifically to comply with new export controls issued by the Trump administration. The company's statement, made in response to media inquiries following the access disruption, marks the first time Anthropic has publicly characterized the shutdown as a compliance action rather than a technical incident.
The move represents a significant development in how the US government is choosing to regulate frontier AI, and how leading AI developers are responding to that regulatory pressure.
Anthropic's Official Position
Where earlier reporting focused on the government's characterization of the models as a national security concern, Anthropic's own framing in its public statement is notably different. The company positioned the takedown as a voluntary compliance action taken in good faith, not as something it contested.
A representative of Anthropic's policy team stated the company is "committed to working within legal frameworks and to supporting responsible governance of AI technology." The statement acknowledged the models were taken offline and indicated that Anthropic is actively engaging with the Commerce Department's Bureau of Industry and Security (BIS) to establish a compliant pathway for restoring access to authorized users.
This posture — compliance first, licensing negotiation second — contrasts with how some technology companies have historically responded to export control actions, where legal challenge is often the first move.
Scope of the Shutdown
Anthropic confirmed the shutdown affects API access to both models globally. The company indicated that the export control order's extraterritorial reach — combined with the difficulty of verifying the nationality of API users at the point of access — made a global suspension the most straightforward path to compliance, pending a formal licensing or carve-out arrangement.
The two affected models are:
| Model | Primary Use Case | Status |
|---|---|---|
| Fable 5 | Advanced reasoning, code generation, agentic workflows | Offline |
| Mythos 5 | Long-context analysis, technical document processing | Offline |
Models in the Claude 4 family — including Claude Sonnet 4.6 and Haiku 4.5 — remain available and are not subject to the current export control order.
Why These Models Specifically
The Trump administration's export control action targeted Fable 5 and Mythos 5 based on capability thresholds that BIS has been developing for AI systems under its ongoing AI regulatory rulemaking. The capability assessment framework evaluates models against criteria in several sensitive domains, including:
- Autonomous technical task completion without human supervision
- Capabilities with potential weapons of mass destruction (WMD) applications, particularly in chemistry and biology
- Cybersecurity offensive capabilities — specifically the ability to autonomously identify, exploit, or weaponize software vulnerabilities
- Dual-use potential in critical infrastructure disruption
Anthropic's Fable 5, in particular, was assessed by BIS as crossing several of these thresholds based on its performance on benchmarks related to autonomous code execution and technical reasoning.
The Policy Debate
The export control action on Anthropic's AI models is part of a broader US government effort to treat frontier AI as a controlled technology analogous to advanced semiconductors, encryption, and certain defense technologies — a framework that has been debated actively within the AI policy community.
Proponents of the controls argue that sufficiently capable AI models represent genuine national security risks if transferred to adversary nation-states, and that the US government has both the authority and the obligation to apply existing export control law to this new technology category.
Critics argue that the current approach is both ineffective and counterproductive: adversary nations are developing their own frontier AI capabilities independently, controls based on API access do not prevent model weights from being replicated or fine-tuned by actors who have already accessed earlier models, and the controls disproportionately harm allied researchers and civil society organizations.
SecurityWeek notes that this is the first confirmed instance of the BIS applying export controls specifically to a commercially available large language model's API access, making the Anthropic case a potential precedent-setter for how the US government treats the AI industry going forward.
What Anthropic's Response Signals
The company's decision to characterize the shutdown as willing compliance — rather than contestation — may reflect strategic positioning ahead of licensing negotiations with BIS. Companies that publicly oppose export control orders often face harder negotiations and slower resolution of licensing pathways.
It may also reflect the reality that Anthropic, as a company focused on AI safety and responsible development, finds it difficult to oppose the basic principle of government oversight of highly capable AI systems, even if it disagrees with specific implementation details.
The path forward likely involves a validated access framework — similar to how high-performance computing export licenses work — in which Anthropic can verify the nationality and use case of API users and provide access to vetted, non-adversary users outside the United States. Building such a framework takes time and requires coordination with BIS.
In the interim, Anthropic has committed to working with affected enterprise customers to provide transition support, including migration to currently unaffected models in the Claude family.