Official Response: US Government Directive to Suspend Fable 5 and Mythos 5

Date: June 12, 2026

Anthropic has received a formal export control directive from the United States government. Invoking national security authorities, the government has ordered the immediate suspension of access to Fable 5 and Mythos 5 for all foreign nationals. This restriction applies regardless of whether the individuals are located inside or outside the U.S. and extends even to foreign national employees within Anthropic.

To ensure full legal compliance with this order, we have been forced to abruptly disable both Fable 5 and Mythos 5 for our entire customer base.

Timeline and Context

The directive was delivered to us today, June 12, at 5:21pm (ET). While the government cited national security as the primary driver, the official correspondence did not provide specific details regarding the exact nature of the concern.

Based on our current understanding, the government believes a method has been discovered to bypass—or "jailbreak"—the safeguards of Fable 5.

Technical Analysis of the "Jailbreak"

Anthropic has reviewed a demonstration of the technique in question. Our findings indicate the following:

Limited Scope: The technique was used to identify a small number of minor, previously known vulnerabilities.
No Unique Uplift: These vulnerabilities are relatively simple. We have confirmed that other publicly available AI models can identify these same flaws without needing any bypass or jailbreak.
Existing Capabilities: The specific "jailbreak" shared with the government essentially involves asking the model to analyze a codebase and suggest fixes for software flaws—a capability already widely available in other models (such as GPT-5.5) and used daily by cybersecurity defenders to secure systems.

Our Approach to Model Safety

Anthropic remains committed to the safety posture detailed in our launch documentation. We have implemented rigorous safeguards to minimize the risk of Fable being used for malicious cybersecurity tasks. In fact, these protections are so stringent that some users have reported them as being overly restrictive.

Prior to launch, we invested thousands of hours in red-teaming, collaborating with:

The US Government
The UK AISI
Various internal teams
Multiple independent third-party organizations

These efforts proved that Fable’s safeguards are significantly more robust than any model previously released. To date, no tester has discovered a universal jailbreak—a method capable of broadly disabling safeguards to unlock a wide array of cyber capabilities.

The "Defense in Depth" Strategy

We acknowledge that absolute, 100% resistance to jailbreaking is likely impossible for any provider. Because non-universal jailbreaks (which elicit specific information in narrow contexts) are an industry-wide vulnerability, Anthropic employs a defense in depth strategy:

Key components of this strategy include:

Increasing Cost: Making universal jailbreaks prohibitively expensive to produce and non-universal jailbreaks narrow in scope.
Active Monitoring: Implementing thorough oversight to detect and neutralize attacks in real-time.
Data Retention: Requiring a 30-day retention period for customer data. While this policy is costly and sometimes unpopular with users, it is essential for researching and mitigating new jailbreak methods.

Conclusion and Path Forward

We have not seen any evidence of a non-universal jailbreak leading to a harmful outcome; the disclosures we have received are either benign or provide no specific advantage.

While we are complying with the legal directive to remove access to Fable 5 and Mythos 5, we strongly disagree with the government's reasoning. We believe that the discovery of a narrow, non-universal jailbreak should not justify the recall of a commercial model used by hundreds of millions of people. If this becomes the industry standard, it would effectively freeze the deployment of all future frontier models.

Anthropic supports the government's ability to block truly unsafe deployments, provided that such actions are part of a transparent, fair, and statutory process grounded in technical evidence.

We believe this current situation is the result of a misunderstanding and are working diligently to restore access for our users as quickly as possible.