The model was pulled on June 12 because real-time nationality verification wasn't feasible under the export control directive. With the Commerce Department lifting those controls, Anthropic redeployed it on July 1, but not without changes. New safety classifiers now intercept requests in the blocked categories, and flagged requests fall back to Claude Opus 4.8 rather than returning an error — a cleaner user experience than a hard failure. For Copilot specifically, enabling Fable 5 requires data retention for the classifiers to function. Pricing sits at $10 per million input tokens and $50 per million output tokens. If your organization disabled it during the suspension period, it won't re-enable automatically — check your Copilot model policy settings.