Bartz v. Anthropic PBC: A Hint at the Direction of the AI Fair Use QuestionDecember 16, 2025

Bartz v. Anthropic PBC: A Hint at the Direction of the AI Faire Use Question

New light is being shed on what use of copyright protected works by artificial intelligence developers may constitute infringement. Anthropic PBC, the artificial intelligence company behind the Claude chatbot, has agreed to pay $1.5 billion to settle a sweeping class action brought by authors and copyright holders who alleged that their books were used without authorization to train the company’s AI models. The proposed settlement, filed in the San Diego U.S. District Court for the Northern District of California, marks the largest monetary settlement to date in a lawsuit over AI training data.

The case, Bartz v. Anthropic PBC (No. 3:24-cv-05417), accused Anthropic of illegally downloading millions of books and incorporating them into its training datasets. The plaintiffs in the Anthropic litigation claim that the creation of these training datasets necessarily required wholesale copying of the works and thus amounts to infringement, whereas Anthropic maintained that any use of legally obtained materials qualified as fair-use under U.S. Copyright Act. Anthropic maintains that its use of the authors’ works does not subject it to liability for copyright infringement and denies any wrongdoing.

Settlement Terms

Under the proposed settlement, Anthropic would create a $1.5 billion fund to compensate writers, publishers, and other rights holders whose works were allegedly included in the disputed dataset. According to the filed proposal, payouts are expected to average about $3,000 per eligible work, though actual amounts will depend on the number of claims filed and administrative costs, with the amount to be split dependent upon the authors and publishers involved with the particular claim.

The company has also agreed to delete all copies of the pirated materials and to certify compliance under court supervision.

What Authors Should Know

Authors and publishers who believe their works may be included can check the official website, AnthropicCopyrightSettlement.com, for eligibility details and a searchable list of affected titles. Rights holders must submit a claim before the March 2026 deadline to receive compensation. However, it may be advantageous for certain rights holders to pursue their own actions. Those wishing to pursue their own lawsuits must opt out by January 7.

Key deadlines include:

• Opt-out deadline: January 7, 2026
• Claim submission deadline: March 23, 2026
• Final fairness hearing: April 23, 2026, in the U.S. District Court for the Northern District of California

If approved, the settlement would release Anthropic from most copyright-related claims connected to the disputed dataset, though it would not prevent future litigation over other AI training practices.

The case is being closely watched by the publishing industry and other tech firms, including OpenAI, Meta, and Google, which face similar lawsuits over alleged misuse of copyrighted content. While settlements like this one are not binding precedent, they set powerful economic and practical benchmarks for future disputes.

That said, the District Court for the Northern District of California issued a binding summary judgment order in Anthropic’s favor wherein it held that use of lawfully acquired books for AI training purposes is transformative and therefore protected by fair-use. However, in ruling on a cross motion, the court did find that the creation and retention of a central library which contained at least some unlawfully acquired works was not transformative and constituted infringement. In other words, the court took the position that the retention of works not lawfully acquired did infringe while lawfully acquired works used in furtherance of training did not. The court did not answer the question of whether unlawfully obtained works used in training artificial intelligence was fair-use which may have disturbed precedent in other federal circuits.

Notably, the holdings in the Anthropic litigation are distinguishable from the landmark Authors Guild v. Google, Inc. (No. 13-4829) decision wherein Google’s creation of a searchable index from the mass digitization of lawfully acquired books was found to constitute fair-use. The Second Circuit found that Google’s use of the works provided research and indexing functionality but did not create a substitution for the original works and created no cognizable harm to the market for books used.

Andrew Morgan is a Trademark Attorney and Chair of the Trademark & Copyright Practice Group at McKee, Voorhees & Sease, PLC (MVS).  For additional information please visit www.ipmvs.com or contact Andrew directly via email at andrew.morgan@ipmvs.com.