Anthropic warns old prompts degrade Claude Fable 5

This update underscores a paradigm shift in prompt engineering where less is more for highly capable models.

• –Overly prescriptive prompting, once a necessity, now appears to constrain or confuse advanced models like Claude Fable 5.
• –The fact that asking the model to "show your reasoning" triggers a refusal suggests Anthropic is tightly controlling how the model exposes its internal chain of thought, likely for safety or alignment reasons.
• –Developers will need to continuously audit and update their prompt libraries, as best practices are becoming rapidly obsolete and model-specific.