Executive Summary
Access to Anthropic's high-end model, Claude Fable 5, has been restored on the AI Gateway platform following the lifting of US Government export controls. The model is unchanged, but it now operates with updated, more robust safety classifiers. To ensure service continuity when these new classifiers are triggered, developers are required to implement a model fallback system, which automatically reroutes requests to alternative models like Opus 4.8 or Sonnet 5.
Key Takeaways
* Restored Access: The Claude Fable 5 model is once again available for use via AI Gateway.
* Updated Safety Classifiers: The model now includes newer, more stringent safety filters, which may temporarily flag routine tasks like coding and debugging.
* Mandatory Model Fallbacks: To prevent service disruptions, users must configure a sequence of fallback models that will be called if a request to Fable 5 is refused by the safety system.
* Data Retention Policy: The model does not support Zero Data Retention. Prompts and completions are stored for 30 days for misuse detection but are not used to train Anthropic's models.
Strategic Importance
This move demonstrates Anthropic's strategy to comply with regulatory and safety demands while ensuring developers retain access to its most powerful models. The mandatory fallback mechanism is a practical solution to maintain high service availability amidst more aggressive safety filtering.