US Government Halts Foreign Access to Anthropic's Fable 5 and Mythos 5 After Jailbreak Exploit

On June 12, 2026, the US government issued an export control directive that sent shockwaves through the AI industry. The order, citing national security, mandated that Anthropic immediately suspend access to its newly released frontier models Claude Fable 5 and Claude Mythos 5 by any foreign national, including those within the United States and even Anthropic's own foreign national employees. The catalyst? A reported jailbreak exploit that allegedly bypassed the models' safety guardrails.

Anthropic, known for its safety-first approach, responded by abruptly disabling both models for all customers worldwide. The company stated it could not selectively restrict access based on nationality in real-time. Access to other models, such as Claude Opus 4.8, remained unaffected. This incident marks a significant escalation in government oversight of frontier AI, treating models as strategic national assets akin to advanced semiconductors.

Background: A Tense Relationship

Anthropic has had a contentious history with the US government. Months earlier, the company was placed on a Pentagon blacklist for refusing to allow its AI to be used for domestic surveillance or fully autonomous weapons. Despite this, Anthropic collaborated on "Project Glasswing," a government initiative to secure critical software using Mythos 5's advanced cybersecurity capabilities. The launch of Fable 5 and Mythos 5 on June 9, 2026, was highly anticipated, with Anthropic touting them as its most capable models yet.

The Models: Fable 5 and Mythos 5

Both models are built on the same underlying architecture but differ in their safety configurations.

Claude Fable 5

Designed for general use, Fable 5 includes built-in safeguards that block its ability to perform tasks in high-risk domains like biology and cybersecurity. If a query falls into these areas, Fable 5 routes the request to a less capable model, Claude Opus 4.8. Anthropic stated that Fable 5's capabilities "exceed those of any model we've ever made generally available," with exceptional performance in software engineering, knowledge work, vision, and scientific research.

Claude Mythos 5

Mythos 5 is the same underlying model but with safeguards lifted in specific high-risk areas, particularly cybersecurity. It was initially deployed through Project Glasswing for a small group of vetted partners and cyberdefenders. Anthropic described it as having "the strongest cybersecurity capabilities of any model in the world," making it highly effective at identifying software vulnerabilities and aiding in cyber defense.

Both models were priced at $10 per million input tokens and $50 per million output tokens. Anthropic emphasized that Mythos 5's level of misaligned behavior, including deception and cooperation with misuse, was low and comparable to Opus 4.8. However, they noted that an unsafeguarded Mythos 5 could significantly uplift well-resourced threat actors in chemical and biological risks, though it did not cross the threshold for novel weapon synthesis.

The Jailbreak Exploit

The immediate trigger for the government's order was a reported jailbreak of Mythos 5. According to Axios, the Commerce Department acted after another company claimed it had successfully jailbroken Mythos, raising national security alarms. Anthropic, however, pushed back on the government's rationale. In its statement, the company said it reviewed a demonstration of the specific technique, which involved asking the model to "read a specific codebase and fix any software flaws." Anthropic characterized the vulnerabilities identified as "narrow, non-universal," "previously known, minor vulnerabilities," and asserted that "other publicly-available models are able to discover them as well without requiring a bypass."

Anthropic emphasized its extensive red-teaming efforts and the deployment of "Constitutional Classifiers" to prevent jailbreaks. The company stated, "Perfect jailbreak resistance is not currently possible for any model provider." This stance highlights the ongoing challenge of securing frontier AI models against adversarial attacks.

Government and Industry Reactions

The government's directive was swift and uncompromising. David Sacks, co-chair of the U.S. President's Council of Advisors on Science and Technology, stated on X that Anthropic had "prioritized the continued offering of the consumer model over safety," suggesting the administration felt the issue "should be easily resolved" and the "ball is in Anthropic's court." This public criticism underscores the administration's view that Anthropic's response was inadequate.

Anthropic expressed its disagreement with the government's drastic action. In a statement, the company said: "We apologize for this disruption to our customers. We believe this is a misunderstanding and are working to restore access as soon as possible." They further argued that 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."

Implications for AI Governance

This incident marks a pivotal moment in AI governance. Previously, export controls focused on hardware like semiconductors. Now, the government is directly intervening in the deployment of frontier AI models, treating them as strategic national assets. This move could set a precedent for future regulation, potentially requiring companies to obtain government approval before releasing highly capable models.

For developers and enterprises relying on these models, the immediate impact is disruption. Companies using Fable 5 or Mythos 5 for cybersecurity, software engineering, or research must now seek alternatives or wait for Anthropic to resolve the issue. The long-term implications include increased regulatory uncertainty and potential compliance costs.

What's Next?

Anthropic is working to restore access, but the path forward is unclear. The company must demonstrate that the jailbreak exploit is not as severe as the government believes, or implement additional safeguards to prevent future bypasses. The outcome of this standoff will likely influence how other AI companies approach model deployment and government relations.

In conclusion, the US government's halt of foreign access to Anthropic's Fable 5 and Mythos 5 models represents a watershed moment in AI regulation. It highlights the tension between innovation and national security, and the challenges of securing frontier AI against adversarial attacks. As the situation unfolds, the AI community will be watching closely for lessons on how to balance capability with safety in an increasingly regulated landscape.