Claude is getting ambitious with watermarking, and I can smell the problems from a mile away

Anthropic is trying to solve the problem of identifying AI-generated text, but a persistent watermark could create another one if AI assistance is mistaken for AI authorship.