A late Friday directive from US authorities has forced Anthropic to abruptly cut global access to its most advanced AI systems, marking one of the most aggressive government interventions yet in frontier model deployment and intensifying the debate over AI security, sovereignty, and export controls.
The order targets Fable 5 and Mythos 5, two of Anthropic’s most capable AI models, and requires the company to suspend access for all foreign nationals, including employees, regardless of location. Anthropic says compliance leaves it no choice but to disable the systems entirely for customers worldwide.
“The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance,” Anthropic said in its statement.
The US government cited national security concerns, but has not publicly detailed the specific risks. Anthropic says it was informed at 5:21 pm ET on June 12, 2026.
Key Highlights
- US government ordered export restrictions blocking foreign national access to Fable 5 and Mythos 5
- Anthropic disabled both models globally to comply with the directive
- Government allegedly flagged a potential jailbreak method targeting Fable 5
- Anthropic disputes severity, calling the issue a narrow, non-universal jailbreak
- Move triggers global concern over AI sovereignty and US tech control
- Access to other Anthropic models remains unaffected
A Sudden Shutdown of Frontier AI Systems
The directive effectively removes access to what Anthropic describes as its most advanced AI systems. Fable 5, a safety-hardened variant of Mythos 5, was only recently released for general use after being held back for limited partner testing due to concerns over cybersecurity capabilities.
According to Anthropic, the government believes it identified a method to bypass Fable 5 safeguards, commonly referred to as “jailbreaking.”
“Our understanding is that the government believes it has become aware of a method of bypassing, or ‘jailbreaking’ Fable 5,” the company said.
Anthropic says it reviewed the demonstration and found only “a small number of previously known, minor vulnerabilities.”
“These vulnerabilities all appear relatively simple, and we have found that other publicly-available models are able to discover them as well without requiring a bypass,” the company said.
What Triggered the US Government Action
The Commerce Department issued the directive under national security authorities. While officials have not publicly elaborated, multiple reports indicate concern about advanced cybersecurity capabilities embedded in Mythos-class systems.
One Axios report cited by news organizations suggests the government may require a license for the “export, re-export or domestic transfer” of the models.
Anthropic says it has only received verbal evidence of what it characterizes as a narrow vulnerability involving software analysis tasks.
“To date, the government has only given us verbal evidence of a potential narrow, non-universal jailbreak,” Anthropic said.
The company also claims it has not been shown evidence of a harmful or system-wide exploit.
Anthropic Defends its Safety Approach
Anthropic strongly rejected the idea that the reported issue justifies a full recall of deployed systems.
The company outlined its safety framework for Fable 5, emphasizing layered defenses, extensive red-teaming, and monitoring.
Anthropic stated:
- It worked with US and UK government groups and third-party evaluators for thousands of hours
- No testers identified a universal jailbreak capable of broadly bypassing safeguards
- Fable 5 safety systems were rated more effective than prior models
- Jailbreak resistance is not perfect for any current AI system
“We suspect that perfect jailbreak resistance is not currently possible for any model provider,” the company said.
Anthropic added that it uses a “defense in depth” strategy designed to make jailbreaks either narrow in scope or expensive to develop, combined with monitoring systems intended to detect misuse quickly.
The company also acknowledged a controversial operational requirement: 30-day customer data retention, which it says is necessary to investigate jailbreak attempts.
Dispute Over Severity of the Vulnerability
Anthropic argues the government’s concern is based on limited evidence and does not reflect a meaningful new risk.
“We have not even received a disclosure of a concerning non-universal potential jailbreak that led to a harmful result,” the company said.
It further stated that the reported capability resembles routine cybersecurity workflows.
“Our understanding is that one potential jailbreak was shared with the government,” Anthropic said, adding that the technique involves asking a model to analyze a codebase and fix software flaws.
The company also said it reviewed the underlying report and believes similar capabilities exist in other leading models, including systems widely used in defensive security work.
Global Access Cut and Industry Shockwaves
The export control order extends beyond customers, affecting foreign national employees inside Anthropic itself, effectively restricting internal access to the models.
CNN reported that this marks one of the most far-reaching US actions yet tied to AI system capabilities, describing it as part of a broader escalation in government scrutiny of frontier AI systems.
The move also lands amid existing tension between Anthropic and the US government, including prior designations labeling the company a “supply chain risk” and ongoing legal disputes over defense-related usage of its tools.
Europe Reacts: “Technological Sovereignty” Debate Intensifies
European officials reacted sharply, framing the decision as a warning about dependence on US-controlled AI infrastructure.
European Commission spokesperson Thomas Regnier said the Anthropic case “further underlines Europe’s need for technological sovereignty,” a view echoed across the continent.
Bruno Retailleau, former French interior minister, described the situation as a wake-up call: “Washington’s decision to cut access to Anthropic’s most powerful models should serve as a wake-up call. In the race for artificial intelligence, a nation that depends on others for its technology is a nation that can be unplugged overnight.”
Other European policymakers emphasized that AI systems are increasingly treated as critical infrastructure, similar to electricity or internet services, and argued that dependence on foreign models creates strategic risk.
The reaction highlights a growing policy divide: AI systems are no longer viewed solely as software products, but as geopolitical assets.
Al Carns, a British MP and former armed forces minister, said the decision highlights how AI systems are now tightly linked to national sovereignty, warning that restricting access to frontier models could reshape how governments think about security, industry dependence, and defense capability.
Benjamin Haddad, French minister delegate for Europe, said the move reflects an accelerating geopolitical struggle over AI, warning that Europe cannot remain “an open market dependent on technologies designed, funded, and controlled elsewhere.”
Jordan Bardella, leader of France’s far-right National Rally and a leading presidential contender, said the decision highlights AI as a question of national sovereignty. “Nations that do not quickly develop their own [models] will always depend more and more on the choices of other powers,” he said.
Other members of the European Parliament made similar arguments, calling for accelerated investment in domestic frontier AI systems. Much of the discussion centered on France-based Mistral, seen as the EU’s strongest AI contender, which is reportedly in talks to raise €3 billion at a €20 billion valuation.
Industry Implications: A New Era of AI Export Controls
The decision to suspend Fable 5 and Mythos 5 access signals a potential shift toward stricter export regimes for advanced AI systems.
If applied broadly, Anthropic warned, the standard could halt frontier model deployment across the industry.
“If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers,” the company said.
The US government has recently moved toward voluntary pre-deployment reporting frameworks for high-capability models, particularly those with advanced cybersecurity potential.
This case suggests those frameworks may be tightening into enforceable controls.
What’s Next?
Anthropic says it is working to restore access as quickly as possible and disputes the interpretation that a narrow jailbreak should trigger full withdrawal from global markets.
“We are complying with the government’s legal directive… However, we disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people,” the company said.
For now, access to Fable 5 and Mythos 5 remains suspended worldwide, while other Anthropic models continue operating normally.
The broader industry is left with an unresolved question: whether frontier AI systems can be governed through traditional export control frameworks, or whether entirely new regulatory models will be required for systems that can cross borders instantly, but are increasingly treated like strategic national infrastructure.
