Anthropic

Type: Lab (defendant in Bartz v. Anthropic)

On this rung Anthropic appears as the defendant in the class action that produced the first court-set price for training data. It downloaded about 7 million book copies from the shadow libraries LibGen (June 2021) and PiLiMi (July 2022), and agreed to a $1.5 billion class settlement covering roughly 500,000 qualifying titles — about $3,000 per title — with preliminary approval on 25 September 2025 (Authors Guild summary).

The case split in a way that matters more than the headline. Training on legally acquired books was held to be fair use; Anthropic was denied summary judgment on the piracy claims, and the settlement resolved those. What Anthropic paid for was how the books were obtained, not what it did with them.

Key Contributions

• The first per-unit price for training data in this KB: ~$3,000 per title, as a penalty for unlicensed acquisition (Authors Guild)
• A ruling that separated acquisition from training — see acquisition versus training

Evidence caveat

Held via the Authors Guild's summary, not the settlement agreement or the rulings. Final-approval status is unresolved in this source (hearing listed for 14 May 2026).

Mentioned In

• The price of training data
• Acquisition versus training
• Bartz v. Anthropic — the source

Related Entities

• The Authors Guild — author of this rung's summary of the settlement
• Reddit — the negotiated side: data bought through a licence rather than paid for through a lawsuit
• Anthropic on the models rung
• Anthropic on the inference-economics rung