On 9 June 2026, Anthropic released Claude Fable 5 — the most capable AI model it had ever made available to the public. Within 72 hours, the US government had ordered it offline. What happened in between is one of the most dramatic episodes in the short history of frontier AI: a landmark launch, a contested jailbreak claim, and a government directive that Anthropic says could halt AI development across the entire industry.

What is Claude Fable 5, and why did it matter?

Fable 5 is not simply a new Claude model. It belongs to a different tier entirely. Anthropic classifies it as a Mythos-class model — a category above its established Opus line, previously accessible only to a small group of cybersecurity partners and critical infrastructure providers through Project Glasswing, a restricted programme run in collaboration with the US government.

Making Mythos-class capabilities publicly available was a deliberate and carefully considered step. Anthropic described Fable 5 as state-of-the-art on nearly all tested benchmarks, with exceptional performance across software engineering, knowledge work, vision, and scientific research. The model’s lead over previous Claude versions grew the longer and more complex the task — a meaningful distinction as enterprises increasingly deploy AI for multi-step autonomous work.

The practical implications were immediate. Stripe reported that Fable 5 compressed months of engineering work into days, performing a codebase-wide migration across a 50-million-line Ruby codebase in a single day — a task that would otherwise have taken a full team more than two months. Cursor’s CEO Michael Truell described it as having “opened up a class of long-horizon problems that were out of reach for earlier models.”

How did Anthropic make such a powerful model safe to release?

The answer was a new safeguard architecture built around safety classifiers — separate AI systems that monitor incoming requests and intercept those flagged as high-risk. Queries related to cybersecurity, biology and chemistry, or model distillation do not receive a Fable 5 response. Instead, they fall back automatically to Claude Opus 4.8, with the user notified. Anthropic said the classifiers trigger in fewer than 5% of sessions on average, meaning the vast majority of users would experience Fable 5’s full capability uninterrupted.

The classifiers underwent extensive red-teaming before launch — including an external bug bounty that produced no universal jailbreaks across more than 1,000 hours of testing. Anthropic acknowledged that perfect jailbreak resistance was likely impossible, and adopted a defence-in-depth strategy instead: make any successful bypass narrow, expensive, and detectable through mandatory 30-day data retention on all Fable 5 and Mythos 5 traffic.

Pricing was set at $10 per million input tokens and $50 per million output tokens — less than half the cost of Claude Mythos Preview. Subscribers on Pro, Max, Team, and Enterprise plans had access included at no extra charge through 22 June.

Why did the US government suspend Claude Fable 5?

Three days after launch, at 5:21pm Eastern Time on 12 June, Anthropic received a government directive. The order, issued under national security authorities, suspended all access to Fable 5 and Mythos 5 by any foreign national — whether inside or outside the United States, including Anthropic’s own foreign national employees. The practical effect was total: to comply, Anthropic had to disable both models for all customers globally. Every other Anthropic model remained unaffected.

The government’s stated concern was a potential jailbreak. The directive provided no specific technical details. Anthropic’s account of what it was shown, however, told a different story. The alleged bypass amounted to asking the model to read a codebase and identify software flaws — a task Anthropic said was well within the capabilities of publicly available models including OpenAI’s GPT-5.5, and one performed routinely by cybersecurity defenders. The vulnerabilities identified were previously known and described by Anthropic as minor.

What does Anthropic’s public pushback mean for AI governance?

Anthropic said it was complying with the legal directive. But it did not stay silent. In a public statement, the company made its position explicit: a narrow, non-universal jailbreak that produces no Mythos-specific uplift should not be sufficient grounds to recall a commercial model deployed to hundreds of millions of people. If it were, Anthropic argued, the same standard applied consistently across the industry would effectively halt all new frontier model deployments.

The statement pointedly noted that the government had only provided verbal evidence of the potential jailbreak, and that Anthropic had validated the capability displayed was already available from other deployed models. The company said it believed the directive was a misunderstanding and was working to restore access.

The episode carries implications well beyond Anthropic. For users and enterprises across Asia and the rest of the world, the suspension demonstrated that access to frontier AI is not simply a product decision — it is a geopolitical one. A US export control directive, issued without public process or technical disclosure, was sufficient to cut off a globally deployed model within hours. As AI capabilities advance and governments become more attuned to their strategic implications, the gap between what AI companies can build and who is permitted to use it may become one of the defining tensions in the technology’s development.

Anthropic has said it intends to restore access as soon as possible. For now, Claude Fable 5 — the most capable model the company has ever released publicly — remains offline.