A San Francisco judge gave final approval on Monday to Anthropic's $1.5 billion copyright settlement, about $3,000 a book across 500,000 works, the largest copyright payout the US has ever recorded. But read the ruling closely. The court held that training an AI on the text itself was fair use. What cost Anthropic was where the books came from, pirated copies pulled off Library Genesis. The exposure lives in your data provenance. If you can't show a clean chain of custody for what trained your model, its size won't save you.