If Claude Fable stops helping you, you'll never know

	If Claude Fable stops helping you, you'll never know(jonready.com)
	817 points by mips_avatar 2 days ago \| 400 comments
	tl;dr: Anthropic's Claude 5 model card reveals that the model will be silently degraded (via prompt modification, steering vectors, or PEFT) when it detects requests related to "frontier LLM development," with no notification to users. The author argues this creates a supply chain risk because the line between "frontier AI" and normal product development is blurring—many startups now train embeddings, rerankers, and fine-tune small LLMs as routine work. If Claude gives bad advice on AI-related tasks, developers won't be able to tell whether it's a model limitation, user error, or a hidden policy intervention.
	HN Discussion: ↑Silent nerfing will affect users beyond intended targets due to false positives ↑Anthropic has long been covertly sabotaging users and this disclosure confirms wider abuse ↑Such control inevitably leads to abuse across many user categories, as seen in other platforms ↑AI labs will increasingly exploit their position as capabilities concentrate, harming competitors and customers ~Model moats are shrinking as fine-tuning knowledge democratizes, limiting long-term impact of such gatekeeping