Anthropic restores Claude Fable 5 access and proposes a jailbreak severity framework

Anthropic says access to Claude Fable 5 has been restored globally after the company and the U.S. government resolved a short-lived export-control disruption that had forced a broad suspension of its newest models. In a June 30 post updated July 1, Anthropic said Fable 5 would return to the Claude Platform, Claude.ai, Claude Code, and Claude Cowork, while Mythos 5 access was restored for a set of approved U.S. organizations.

The company framed the episode as both an availability update and a safety process change. Anthropic said the original suspension followed a U.S. directive after officials reviewed a report in which Amazon researchers found a way to bypass Fable 5 safeguards for some cybersecurity tasks. Anthropic said its own testing found the case did not reveal unique Mythos-level cyber capabilities, but it trained an improved classifier to block the specific behavior described in the report.

Why developers and security teams should care

• Fable 5 availability is returning across consumer, developer, and work products, with cloud platform re-enablement planned as quickly as possible.
• For Pro, Max, Team, and select Enterprise plans, Anthropic says Fable 5 is included for up to 50 percent of weekly usage limits through July 7, after which usage credits apply.
• The new classifier is described as blocking the reported bypass in more than 99 percent of cases, though Anthropic warns it may increase false positives in ordinary coding and debugging work.
• Anthropic says it is working with Amazon, Microsoft, Google, and other Glasswing partners on a shared way to score the severity of AI jailbreaks.

The broader story is that frontier AI releases are now entangled with cyber-safety review, export-control interpretation, and enterprise reliability. Anthropic is asking customers to accept a more cautious classifier in exchange for broader model access. It is also trying to turn a public disruption into an industry process for describing jailbreak risk consistently, so future findings can be triaged with clearer signals for vendors, customers, and government partners.

Sources

• www.anthropic.com

Cover image: jurvetson, source, licensed under BY.