Claude Fable 5 access restored on AI Gateway
Claude Fable 5 is back. Here's what changed—and why your safety classifiers matter now.
Why it matters
Anthropic restored access to Claude Fable 5 on AI Gateway after US export control lifting, with upgraded safety classifiers and new model fallback requirements. This affects developers integrating the model into production workflows.
The key facts
13 to knowClaude Fable 5 (Mythos-class) access restored following US Government export control decision
Model availability window: June 9-12, now restored
Safety classifiers updated and more robust than previous version
Routine tasks (coding, debugging) may trigger new safety classifiers
Model fallbacks feature available to ensure request servicing when Fable 5 refuses
30-day data retention policy; prompts/completions not used for training
Zero Data Retention not supported due to cumulative misuse pattern detection requirements
Claude Fable 5 (Mythos-class model) access restored following US Government export control decision
Model was previously available June 9-12; same model now with updated safety classifiers
Safety classifiers triggering on routine tasks like coding and debugging
Model fallback feature available to route requests to alternative models if Fable 5 refuses
Zero Data Retention not supported; 30-day data retention policy for safety monitoring
Fallbacks work across all API formats (REST, streaming, etc.)
Go to the source
Vercel Blogvercel.com
Publisher excerpt: Access to Claude Fable 5, the Mythos-class model, has now been on AI Gateway following the to lift the export controls.restoredUS Government's decision is the same model that was available between June 9 and June 12. What has changed is the safety classifiers, which are now updated and more…