Anthropic Restores Fable 5 Worldwide and Proposes a Jailbreak-Severity Framework

After a US export-control suspension that lasted nearly three weeks, Anthropic returned Fable 5 globally on July 1 and set out a way to score how dangerous a jailbreak really is.

Jul 5, 2026Source: Anthropic

Anthropic Restores Fable 5 Worldwide and Proposes a Jailbreak-Severity Framework

Anthropic returned its Claude Fable 5 model to users worldwide on July 1, 2026, nearly three weeks after the US government imposed export controls that pulled it — and the more specialized Mythos 5 — from general availability. Alongside the restoration, the company set out a proposed industry-wide framework for scoring how serious an AI "jailbreak" actually is.

How the suspension unfolded

Fable 5 and Mythos 5 were released on June 9. On June 12 the US government applied export controls; on June 26 it approved Mythos 5 access for US organizations; on June 30 the controls were lifted; and on July 1 Fable 5 was redeployed globally. Anthropic distinguishes the two models: Fable 5 ships with strong safeguards for general use, while Mythos 5 carries fewer safeguards and is intended only for defensive cybersecurity work.

The jailbreak at the center

The controls followed a report from Amazon researchers, who found that Fable 5, when prompted so that it identified a number of software vulnerabilities, produced code demonstrating how to exploit one of them. Anthropic says its own testing showed the weakness was not unique to its model: Claude Opus 4.8, GPT-5.5 and Kimi K2.7 could identify the same vulnerabilities, and every model tested could produce the same exploitation demonstration. The company says it has since trained an improved safety classifier that blocks the specific behavior in over 99% of cases.

A framework for scoring jailbreaks

To keep incidents like this in proportion, Anthropic proposed judging a jailbreak on four criteria: capability gain (how far beyond existing tools it takes a user), breadth (how many distinct offensive tasks the same technique unlocks), ease of weaponization (how much human effort it takes to turn into a real attack), and discoverability (how easily someone could obtain the technique). The company says it developed the approach with Amazon, Microsoft, Google and other partners, and with the US Commerce Department's Center for AI Standards and Innovation (CAISI).

Closer ties to government

The return came with new commitments to work with government: pre-release access and evaluation of frontier models, faster information-sharing on safeguards, dedicated resources for joint research, and work toward a common industry security standard. Fable 5 is available again on the Claude Platform, Claude.ai, Claude Code and Claude Cowork, with access through AWS, Google Cloud and Microsoft Foundry to be restored as quickly as possible.

Why It Matters

The episode is a live test of how the US government's new frontier-AI review works in practice: a leading model can be pulled from the market and restored within weeks. Anthropic's severity framework is an attempt to keep "jailbreak" alarms proportional to real-world risk — especially when, as here, the same weakness turns up across rival models — and its tighter government commitments show how quickly frontier labs are being pulled into a formal national-security relationship.

Filed underAnthropic AI Safety US AI Regulation #AI security #Claude #jailbreak