English
US Government Restricts Anthropic's Advanced AI Models Over National Security
The United States government has ordered Anthropic to suspend foreign access to its most advanced artificial intelligence models — Claude Fable 5 and Mythos 5 — following a security report from Amazon that revealed vulnerabilities in the systems. The decision, announced on June 12, 2026, marks a significant escalation in Washington's efforts to control the global dissemination of frontier AI technology.
According to sources cited by The Wall Street Journal, the export control directive was triggered after Amazon CEO Andy Jassy shared internal research with White House officials, including Treasury Secretary Scott Bessent. Amazon researchers had successfully demonstrated a method to bypass Fable 5's safety guardrails, prompting the government to act swiftly.
Amazon's Discovery and the 'Jailbreak' Vulnerability
Amazon's cybersecurity team discovered that specific sequences of prompts could cause Fable 5 to reveal information useful for cyberattacks, effectively bypassing the model's built-in safety mechanisms. The researchers reported their findings to Amazon leadership, who then escalated the matter to the highest levels of the US government.
A spokesperson for Amazon confirmed that it is not unusual for governments to consult the company on security risks but declined to share details about the conversations. Anthropic's AI safety measures have been a subject of debate, with the company positioning itself as a leader in responsible AI development.
The vulnerability specifically targeted Fable 5, a version of Anthropic's Mythos model designed for general use. Unlike its more restricted counterpart Mythos 5 — which is released only to trusted cybersecurity researchers through Project Glasswing — Fable 5 was intended for broad commercial deployment.
Immediate Impact: Global Shutdown of Fable 5 and Mythos 5
Anthropic responded by abruptly disabling both models for all users worldwide. The company stated that the US Commerce Department's Bureau of Industry and Security issued an export control directive requiring the suspension of access for any foreign national, both inside and outside the United States. Since Anthropic could not verify user citizenship in real time, it chose to disable the models entirely to ensure compliance.
"We believe the government should have the ability to block unsafe deployments through transparent processes, but this action did not adhere to those principles," an Anthropic spokesperson said in a statement. The company argued that the identified vulnerability was a "narrow, non-universal jailbreak" that could also be reproduced on competing platforms like OpenAI's GPT-5.5.
Anthropic emphasized that Fable 5 had undergone extensive red-team testing with the US government, the UK's AI Safety Institute, and third-party auditors without finding a universal jailbreak. The company warned that pulling models over narrow vulnerabilities could set a problematic precedent for the entire industry.
What Are Fable 5 and Mythos 5?
Anthropic launched Claude Fable 5 and Claude Mythos 5 on June 9, 2026, just days before the government order. Fable 5 is the first publicly available "Mythos-class" model, representing a tier above the previous Claude Opus 4.8. Key features include:
- State-of-the-art performance on benchmarks for software engineering, knowledge work, vision, scientific research, and long-context tasks
- 1 million token context window and 128,000 max output tokens
- Safety routing that redirects sensitive queries to the more conservative Claude Opus 4.8, triggering in under 5% of sessions
- Pricing at $10 per million input tokens and $50 per million output tokens
Mythos 5, the underlying model with fewer safeguards, was released exclusively through Project Glasswing to cybersecurity defenders and research institutions. The growing AI export control debate centers on whether such powerful models should be treated as controlled technologies akin to nuclear materials.
Broader Implications for Europe and Global AI Access
The US export controls have far-reaching consequences, particularly for Europe. European businesses and governments have built much of their AI infrastructure on American models. With access to the most advanced systems now restricted, they face the prospect of relying on outdated or less capable technology.
"This is a stark wake-up call for Europe," a European Commission official told Politico. The move exposes the continent's heavy dependence on US AI models and has sparked urgent calls for technological sovereignty. French AI company Mistral, the EU's most credible contender, is reportedly seeking €3 billion in funding to develop competitive frontier models.
Policy experts warn that the broad restrictions could damage international research collaborations and hamper domestic tech development outside the United States. The future of AI regulation in Europe may accelerate as a direct result of this decision.
Anthropic's Dispute and Industry Reaction
Anthropic has publicly disagreed with the government's assessment. The company noted that the vulnerability was minor and easily discoverable in other publicly available models. Anthropic's statement argued that applying this standard across the industry would halt all new frontier model deployments.
The dispute comes amid existing tensions between Anthropic and the Trump administration. Earlier in 2026, the Department of Defense designated Anthropic a "supply chain risk" after the company refused to remove contractual restrictions prohibiting the use of its AI for domestic surveillance and fully autonomous weapons. Anthropic, which recently filed for a confidential IPO and is valued at approximately $965 billion, is working to restore access to its models.
Industry observers note that the US government's action represents a paradigm shift: applying traditional export control law — originally designed for physical goods — to cloud-hosted AI models accessed via API. This precedent could reshape how frontier AI is governed globally.
FAQ
Why did the US block foreign access to Anthropic's AI models?
The US Commerce Department issued an export control directive after Amazon researchers demonstrated that Fable 5's safety guardrails could be bypassed, potentially enabling cyberattacks. The government cited national security concerns.
Which Anthropic models are affected?
Claude Fable 5 (the general-use version) and Claude Mythos 5 (the restricted cybersecurity version) have been disabled globally. Other Anthropic models, including Claude Opus 4.8, remain unaffected.
How long will the restrictions last?
Anthropic is actively working with the US government to restore access. The company has stated it will share more technical details and is seeking a resolution, but no timeline has been provided.
What does this mean for European AI development?
European companies and governments that rely on US AI models may face significant disruptions. The incident has intensified calls for Europe to develop its own frontier AI capabilities to reduce dependency on American technology.
Is this the first time the US has restricted AI model access?
This is the most direct action targeting AI models themselves. Previous restrictions focused on AI chips and development tools. This marks an escalation in US efforts to control foreign access to advanced AI capabilities.
Sources
- Reuters: US blocks foreign access to Anthropic's most advanced AI models
- CNBC: Anthropic disables access to Fable 5 and Mythos 5
- The Guardian: Anthropic disables advanced AI models after US government order
- Politico EU: US Anthropic order exposes EU AI dependency
- Anthropic: Statement on the US government directive
Follow Discussion