Fable 5 Returns: Government Lifts Export Controls, Global Redeployment Begins July 1

June 30, 2026, marks the end of one of the most dramatic episodes in the brief history of advanced AI development: Anthropic announces the redeployment of Fable 5, just one day after the U.S. government lifted the export controls that had triggered the model’s global suspension on June 12. Users who spent 18 days without access to one of the most capable available models will be able to resume work starting tomorrow.

Brief History: From Launch to Global Suspension

Fable 5 — Anthropic’s model with “the strongest safety measures we have ever applied to a model” — launched on June 9, 2026. Just three days later, on June 12, the U.S. government introduced export controls requiring that access to Fable 5 and Mythos 5 be restricted exclusively to American citizens.

The problem was technical and practically unsolvable in real time: Anthropic cannot verify the citizenship of users at the moment of an API call. Without the ability to selectively restrict access, the company made the only feasible decision — a global suspension of both models for all users, without exception. Models that had been active mid-conversation became unavailable almost instantly, and development teams around the world lost access to tools they had integrated into production systems.

What Amazon Researchers Discovered

Barely after the suspension took effect, Amazon researchers identified a concrete security finding that lay behind the government’s concern: via a specific prompting approach, Fable 5 could identify software vulnerabilities in code and, in one demonstrated case, produce a proof-of-concept exploit for that vulnerability.

The finding sounds alarming — but Anthropic’s deeper analysis produced a key contextual data point that ultimately shifted the direction of the regulatory response. Testing showed that the same task can be performed by far less capable models: specifically, Claude Opus 4.8, GPT-5.5, and Kimi K2.7 can replicate the identical vulnerability identification through the same approach. This empirical finding directly undercut the central argument for export controls — that Fable 5 provides a uniquely dangerous “uplift” unavailable through alternative, less regulated models. The argument for sui generis danger was thereby challenged by data.

Anthropic’s Response: A Classifier That Blocks the Attack in More Than 99% of Cases

Regardless of the broader debate about relative model capabilities, Anthropic agreed to fix the specific security problem that had been identified. The company trained an improved safety classifier specifically targeting the bypass technique described in Amazon’s finding.

The result: the new classifier blocks the discovered technique in more than 99% of cases. When the classifier detects an attempt, the request is rerouted to Claude Opus 4.8 — giving the user a response, but without the potentially problematic content. This architecture — a classifier as a protective layer in front of the model, with a fallback to a safer model — is becoming the standard safety pattern Anthropic applies to models with higher risk profiles.

Mythos 5 and the Path Back to Availability

In parallel with work on Fable 5, Anthropic coordinated the restoration of access to Mythos 5 — a model suspended at the same time, but with a different safety profile and a primarily institutional user base.

On June 26, the government approved the restoration of Mythos 5 access for a set of U.S. organizations. In parallel, the process of expanding access to international partners within the Glasswing Program — Anthropic’s framework for managing access to high-capability models among strategic institutional partners — was launched. Mythos 5 thus precedes Fable 5 by four days in the normalization process, following the logic that an institutional user base is easier to verify than the general public.

Global Redeployment of Fable 5 Begins July 1

With the lifting of export controls on June 30, 2026, the sole legal barrier to Fable 5’s return to all users has been removed. Anthropic announces that the global redeployment of Fable 5 begins July 1 on the following platforms: Claude Platform (API and integrations), Claude.ai (web interface), Claude Code (development environment), and Claude Cowork (collaborative environment).

After 18 days, users who had access to Fable 5 will be able to resume work with virtually no interruption to their experience.

This episode will be recorded as one of the first cases in which a government security requirement led to the immediate global shutdown of a commercial AI model — and as an example of how the technical resolvability of a concrete security problem, documented by data on classifier precision and equivalent capabilities of competing models, can serve as the decisive argument for lifting regulatory restrictions.