Anthropic Restores Access to Claude Fable 5 After US Export Controls Lifted

On July 1, Anthropic announced it had restored Claude Fable 5 to users worldwide. The move came after US export controls, which had suspended access since June 12, were lifted. Those controls were triggered by a report from Amazon researchers who found a way to bypass Fable 5's safeguards, enabling the model to identify software vulnerabilities and demonstrate exploit code.

Timeline and Resolution

The US government imposed the export controls on June 12, citing concerns that Fable 5 and its less-safeguarded counterpart, Mythos 5, could be misused by malicious actors. The order forced Anthropic to restrict access to foreign nationals, which led to a full suspension of both models. As of June 30, the controls were lifted. Fable 5 is now available on the Claude Platform, Claude.ai, Claude Code, and Claude Cowork. For Pro, Max, Team, and select Enterprise plans, Fable 5 is included for up to 50% of weekly usage limits through July 7, after which usage credits apply. Access on AWS, Google Cloud, and Microsoft Foundry is being re-enabled as quickly as possible.

Mythos 5 access has been restored for a set of US organizations following government approval on June 26, with ongoing coordination to expand access to broader Glasswing program partners.

New Safety Classifier

In response to the reported jailbreak, Anthropic trained an improved safety classifier that blocks the specific behavior described in the Amazon report in over 99% of cases. The company stressed that the technique did not expose unique Mythos-level cyber capabilities, and that many other models, including Claude Opus 4.8, GPT-5.5, and Kimi K2.7, could identify the same vulnerabilities and produce similar exploit demonstrations. Anthropic noted that the behavior reflected a borderline case for Fable 5's safeguards, which deliberately block some benign requests to maintain a safety margin. The new classifier comes with an increased rate of false positives during routine coding tasks, which Anthropic says it will continue to refine.

Industry regulators have taken notice. The US Department of Commerce's Center for AI Standards and Innovation (CAISI) tested both the prior and new safeguards and agrees they are extraordinarily strong.

Industry Framework for Jailbreak Severity

Anthropic also proposed a consensus industry framework for assessing AI jailbreak severity, developed with Amazon, Microsoft, Google, and other Glasswing partners. The framework scores jailbreaks on four criteria: capability gain (how much the jailbreak enhances offensive ability beyond existing tools), breadth of capability gain (how many distinct tasks the technique enables), ease of weaponization (how much effort is needed to turn it into an attack), and discoverability (how easy it is to obtain the technique).

The proposal suggests using this severity framework to calibrate responses, with the most severe jailbreaks triggering immediate mitigations. Anthropic also launched a new HackerOne program for security researchers to submit potential cyber jailbreaks in Fable 5.

Deepening Government Collaboration

Anthropic outlined several commitments to strengthen collaboration with the US government, building on nearly two years of pre-existing partnerships. These include pre-release government access and evaluation of models that advance the capability frontier, rapid information sharing on safeguards and jailbreaks, dedicated resources for joint research, and work toward a shared voluntary security and evaluation standard for frontier model providers. The company hopes this collaboration will serve as a template for global coordination on AI risks and benefits, and called for these rules to be codified in strong regulation applied equally across frontier model developers.