The US government has ordered Anthropic to immediately shut down access to two of its AI models, Fable 5 and Mythos 5, citing national security concerns over a potential security vulnerability. The directive, issued on June 12, 2026, requires Anthropic to disable these models for all users worldwide, including the company's own foreign employees, though access to Claude and Anthropic's other models remains unaffected.

What Triggered the Government's Action?

According to Anthropic's statement, the government became aware of a method to bypass, or "jailbreak," Fable 5's safety guardrails. When Anthropic reviewed a demonstration of this technique, they found it exposed only a small number of previously known, minor vulnerabilities. The company noted that these vulnerabilities appear relatively simple and that other publicly available models, including OpenAI's GPT-5.5, can discover them without requiring a bypass.

The specific jailbreak technique involved asking the model to read a codebase and identify software flaws, a capability Anthropic argues is widely available across the AI industry and is used daily by cybersecurity professionals to defend systems. Anthropic stated they have not received disclosure of any jailbreak that led to actual harmful results.

How Does Anthropic's Defense Strategy Compare to Industry Standards?

Anthropic has outlined its approach to securing Fable 5, which the company believes is more robust than previous models. The company's safeguards underwent thousands of hours of red-teaming, a process where security experts attempt to break the model's safety features. This testing involved collaboration with the US government, the UK's AI Safety Institute (AISI), multiple third-party organizations, and Anthropic's internal teams.

• Defense Depth Strategy: Anthropic designed Fable 5 to make jailbreaks either narrow in scope, affecting only specific circumstances, or very expensive to produce if they were universal. The company combined this with thorough monitoring to quickly detect and shut down successful attacks.
• Data Retention Policy: Anthropic requires 30-day retention of customer data for Fable 5, a policy that carries real costs but enables the company to research and mitigate emerging jailbreaks more effectively.
• Safeguard Effectiveness: Anthropic's testing showed that Fable's safeguards are substantially more effective than those of any previously deployed model, though the company acknowledges that perfect jailbreak resistance may not be currently possible for any provider.

Anthropic stated that it stands by this defense-in-depth strategy, arguing that it reduces risks posed by Fable to levels comparable to existing models already deployed across the industry.

What Does Anthropic Say About the Government's Decision?

While Anthropic is complying with the government's legal directive, the company has expressed strong disagreement with the decision. Anthropic argues that finding a narrow, non-universal jailbreak should not be cause for recalling a commercial model deployed to hundreds of millions of people. The company warned that if this standard were applied across the entire AI industry, it would essentially halt all new model deployments for all frontier AI providers.

"We believe the government should have the ability to block unsafe deployments, as part of a statutory process that is transparent, fair, clear, and grounded in technical facts. This action does not adhere to those principles," Anthropic stated.

Anthropic, in statement on government directive

Anthropic has committed to sharing more details about the jailbreak within 24 hours and stated it is working to restore access to Fable 5 and Mythos 5 as soon as possible. The company apologized for the disruption to its customers.

What Are Americans' Broader Concerns About AI Safety?

The government's action comes as Anthropic released findings from its first Anthropic Public Record survey, a new effort to understand how the general public thinks about AI development and safety. The survey included nearly 52,000 Americans and was conducted in November and December of 2025.

When asked what would best ensure AI benefits humanity, Americans ranked two actions as most important. These priorities reflect broader concerns about corporate accountability in AI development:

• Legal Liability: 47% of Americans said holding AI companies legally liable for harm is the highest-leverage action to ensure AI benefits humanity.
• Safety Over Growth: 44% prioritized companies prioritizing safety over growth as the most important step.
• Government Regulation: Over 70% of survey respondents believe the government should play a role in regulating AI, with support for intervention being bipartisan.

The survey also revealed that only 15% of Americans trust AI companies to make decisions about how AI is developed and used, underscoring public skepticism about industry self-regulation.

On specific areas where Americans want government action, privacy protection topped the list at 56%, followed by child safety at 52% and liability for harm at 49%.

How to Stay Informed About AI Safety Developments

• Monitor Official Statements: Follow announcements from major AI companies like Anthropic, OpenAI, and Google about security vulnerabilities and government directives affecting their models.
• Track Regulatory Actions: Keep watch for government directives and regulatory frameworks emerging from agencies like the US government and international bodies such as the UK's AI Safety Institute.
• Review Public Research: Engage with surveys and studies from AI companies about public attitudes toward AI safety, as these often precede regulatory changes and industry standards.

The suspension of Fable 5 and Mythos 5 represents a significant moment in AI governance, highlighting the tension between rapid AI development and government oversight. As AI capabilities advance and adoption deepens, how regulators, companies, and the public navigate questions of safety and accountability will likely shape the industry's future trajectory.