What Authors Need to Know About the Anthropic Settlement (Bartz v. Anthropic, $1.5bn class settlement)
A court-set per-unit price for pirated training data: ~$3,000 per title across ~500,000 qualifying titles from a $1.5bn fund, against 7 million copies from LibGen and PiLiMi. The fair-use ruling covered TRAINING; the $1.5bn covered ACQUISITION — the distinction most coverage collapses. Final-approval date unresolved (hearing 2026-05-14 per this source).
Bartz v. Anthropic — the $1.5bn class settlement
What this is
A class-action settlement over Anthropic's acquisition of books from the shadow libraries LibGen and PiLiMi, summarised for authors by the Authors Guild. It is read here for one thing the data rung has never held: a per-unit price for training data, set by a court rather than by a negotiation.
Terms, as published
| Term | Value |
|---|---|
| Settlement fund | $1.5 billion |
| Approximate qualifying titles | ~500,000 |
| Implied payment per title | ~$3,000 (minimum, before fees and costs) |
| Copies at issue | 7 million book copies |
| Sources of the copies | LibGen (downloaded June 2021) and PiLiMi (July 2022) |
| Eligibility | The work must have an ISBN or ASIN, and a US Copyright Office registration made within five years of publication, and registered either within 3 months of first publication or before Anthropic downloaded it (10 August 2022) |
| Preliminary approval | 25 September 2025 |
| Claims deadline | 30 March 2026 |
| Final approval hearing | 14 May 2026, 2 p.m. PT |
The distinction that matters
Judge Alsup's ruling held that using legally acquired books to train a model was fair use, and separately denied Anthropic summary judgment on the piracy claims. The settlement therefore prices acquisition, not training.
That is the teachable split, and it is the one most coverage collapses:
- Training on a book you bought — adjudicated fair use in this case.
- Obtaining the book from a pirate library — not resolved in Anthropic's favour, and the thing $1.5bn bought peace on.
The Authors Guild also notes the fair-use holding as decided applies to the three named plaintiffs, not to the class at large — so it is not a general licence, and reading it as "training is fair use, settled" overstates it.
What number this moves
data had zero sources. The rung's question is what the input side of the token-price chain costs. This gives it a first anchor:
- ~$3,000 per title, court-imposed, for unlicensed acquisition — the penalty price, which is the ceiling any negotiated licence sits under.
- $1.5bn / 7 million copies ≈ $214 per copy downloaded, if read against copies rather than titles. The two denominators give very different unit prices and the settlement uses titles, not copies; the $214 figure is arithmetic on this page's own numbers and is offered only to show how sensitive the "price of a book" claim is to which denominator is used. Cite the $3,000-per-title figure, not this one.
Limitations and open items
- The final-approval status is unresolved in this source. This page states a final approval hearing of 14 May 2026. Secondary reporting has put final approval in July 2026 — unverified, and deliberately not cited here. Resolve from the docket before any published claim turns on the date.
- The Authors Guild is an advocacy organisation and a party-adjacent commentator, not the court. The terms above are its summary; the operative document is the settlement agreement on the docket.
- "~$3,000 per title" is a minimum before fees and costs, and the actual per-claimant payment depends on how many valid claims are filed against the ~500,000 estimate.
- The eligibility gate (Copyright Office registration within five years, plus ISBN/ASIN) excludes a large share of published work. The settlement prices registered, commercially published books — not "text" in general — which limits how far the unit price generalises to the web-scale corpora that dominate pretraining.
- This is one jurisdiction, one defendant, one pair of shadow libraries. It is a datapoint on the price of unlicensed acquisition, not a market rate.
Standing
I only read about it. No first-person execution here; a published piece leaning on this must be labelled a coverage piece, not a teaching piece, unless it teaches the acquisition-versus-training distinction — which is where the teachable content actually is.
No live market call rests on this rung (gate 3 unmet on data); this is brand/teaching only.
Source: What Authors Need to Know About the Anthropic Settlement — The Authors Guild.