Overview
- Anthropic restored global access to Claude Fable 5 on July 1 after a U.S. export-control pause and redeployed the model with new safety classifiers that the company says flag more harmless prompts than before.
- Microsoft CEO Satya Nadella publicly said on July 16 that Fable’s frequent refusals and editorial controls “don’t make sense,” a rare critique from a major investor and cloud partner.
- Fable 5 uses an automated safeguard that routes requests it deems high‑risk — such as queries about model distillation, offensive cybersecurity, biology, or chemistry — to an older model, Claude Opus 4.8, and some users have reported false positives and redirected outputs.
- Anthropic defends the limits as a way to stop large‑scale model distillation that could let competitors copy years of R&D, but the policy clashes with customers’ need for predictable, usable models and complicates commercial ties with Microsoft.
- The dispute increases pressure on enterprises to adopt in‑house or open‑weight alternatives, raises questions about government pre‑release oversight of frontier models, and could reshape vendor terms, pricing, and who controls advanced AI access.