Breaking

Federal Judge Approves Anthropic's Record $1.5 Billion Copyright Settlement With Authors

July 20, 2026 at 20:17 EDT

Policy & Regulation
Foundation Models

A U.S. federal judge has granted final approval to Anthropic's $1.5 billion settlement with authors and publishers, closing one of the largest copyright cases in American history and setting a marker for how AI companies acquire their training data. On July 20, 2026, Judge Araceli Martinez-Olguin of the U.S. District Court for the Northern District of California signed off on the deal in *Bartz v. Anthropic*, ordering the company to pay a minimum of $1.5 billion to roughly 500,000 works whose owners qualify for compensation of about $3,000 apiece.

July 20, 2026 · Bartz v. Anthropic

A $1.5 Billion Line in the Sand for AI Training Data

A U.S. federal judge granted final approval to Anthropic's record copyright settlement, ordering payment for roughly 500,000 pirated works and drawing a sharp divide between lawfully acquired and stolen data used to train large language models.

$1.5B

Minimum total settlement payout

~500K

Works eligible for compensation

~$3,000

Payment per qualifying work (often split 50/50)

90%+

Reported claims rate among rights holders

Why Anthropic Chose to Settle

Statutory damages can reach far beyond the settled per-work amount — trial risk drove the deal. Same unit: dollars per work, drawn to scale.

~$3,000

settled per work

up to $150,000

statutory max per work at trial

Takeaway: exposure of up to ~50× the settled amount per work pushed Anthropic to resolve rather than face a jury.

The Ruling That Split the Case in Two

Lawfully acquired

Training AI on legitimately obtained books ruled fair use — "exceedingly transformative."

→

Pirated sources

Using books from shadow libraries (LibGen, PiLiMi) held unlawful — headed toward jury trial.

Anthropic also agreed to destroy the infringing copies. The deal grants no license for future use.

Seen as accountability

Some authors and advocates view the payout as meaningful accountability for the use of pirated material — with claims interest above 90%.

Seen as falling short

Critics say the sum is modest, fair use effectively blessed training on legitimate books, and future use isn't prevented. Some rights holders opted out.

Practical, not precedent-setting

Because the matter was resolved by settlement rather than a contested judgment, it won't be appealed and creates no binding precedent — but it elevates the value of licensing markets for AI training data.

Ongoing suits against Google, Meta and OpenAI will test the fair-use boundaries this deal left unresolved.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.