Anthropic Updates Claude Fable 5

Anthropic has officially relaunched its Claude Fable 5 model, making it available to users worldwide from 1 July. The return follows the introduction of new safety measures and revised usage limits after security concerns temporarily delayed the model’s wider rollout.

The announcement was made alongside confirmation from the United States government, which said it had worked closely with Anthropic to review the model before approving its release.

US Secretary of Commerce Howard Lutnick stated that government officials and Anthropic spent several weeks evaluating Claude Fable 5 to ensure it met safety expectations while supporting the country’s broader artificial intelligence strategy.

Although the approval allows the model to become available globally, access to Mythos 5 remains more limited. At present, the model has only been restored for a selected group of organisations within the United States, with broader availability expected at a later stage.

The relaunch comes after a security issue was identified during testing. According to Anthropic, the vulnerability, which was originally reported by Amazon, has now been addressed through the introduction of a significantly improved safety classifier.

A classifier is an automated security system designed to analyse user prompts before they reach the model. Its purpose is to distinguish between legitimate requests and prompts that attempt to bypass the model’s built-in safety protections.

Anthropic says the updated classifier is capable of identifying and blocking around 99% of the reported bypass attempts. When a prompt is flagged, users will receive a notification explaining that their request has been blocked.

Rather than rejecting these requests completely, Anthropic has introduced an alternative process. Blocked prompts will instead be redirected to Claude Opus 4.8, allowing users to continue their work while maintaining stronger security controls.

The company has acknowledged that the new protection system may occasionally identify legitimate requests as potentially unsafe. This could affect some users carrying out routine software development, coding or debugging tasks.

Anthropic says it is continuing to refine the classifier to reduce these false positives while maintaining high levels of protection against misuse.

Alongside the new safeguards, the company has also introduced temporary usage limits for Fable 5. Eligible users on supported Claude subscription plans will initially receive access capped at up to 50% of their normal weekly usage allowance.

These temporary limits will remain in place until 7 July. After that date, access to Claude Fable 5 will move to a usage credit model, allowing customers to purchase additional capacity based on their individual needs.

The model is now available through several Anthropic services, including the Claude Platform, Claude.ai, Claude Code and Claude Cowork. Anthropic also confirmed that support across major cloud providers, including AWS, Google Cloud and Microsoft Foundry, will return as quickly as possible.

Meanwhile, access to Mythos 5 will continue expanding gradually. Anthropic says it is working closely with government agencies to extend availability beyond the initial group of approved US organisations participating in its Glasswing programme.

The relaunch of Claude Fable 5 demonstrates Anthropic’s continued focus on balancing innovation with responsible AI development. By introducing stronger safeguards while maintaining broad access, the company aims to improve user confidence as its latest AI models become available to a wider audience.