Claude Models Refuse Creative Prompts More Often as Anthropic Tightens Safety After Fable 5
Updated
Updated · Android Authority · Jul 8
Claude Models Refuse Creative Prompts More Often as Anthropic Tightens Safety After Fable 5
3 articles · Updated · Android Authority · Jul 8
Summary
Recent Claude models are showing more pushback on fiction, debate and hypothetical prompts, with users reporting refusals even after clearly framing requests as creative or academic.
That caution appears tied to stricter safety guardrails after Fable 5 was accused by the government of posing a security risk, though the report cites no formal Anthropic confirmation.
The problem is also inconsistent: Sonnet 5 refused one religion-themed story prompt in one chat, then completed the same scenario in a fresh conversation.
Newer models such as Opus 4.8, Sonnet 5 and Sonnet 4.6 were described as more restrictive than Opus 4.6 and 4.7, while Fable 5 was said to be especially sensitive on biology questions.
The broader takeaway is a trade-off between Claude’s strong long-context performance and weaker reliability on sensitive prompts, a contrast that may make rivals like Gemini look more consistent.