Five Takeaways from the Bartz et al. v. Anthropic PBC Order on Fair Use

Issue 36 

Two fair use opinions were issued this week in AI copyright infringement cases. The first ruling in Bartz et al. v. Anthropic PBC is the topic of this week’s newsletter. Next week I’ll write about the second fair use ruling in Kadrey et al. v. Meta Platforms, Inc. 

On June 23, 2025, the Hon. William Alsup issued an order on fair use in response to Anthropic’s motion for summary judgment in Bartz et al. v. Anthropic PBC.[i] The order identified and analyzed three distinct uses of the authors' copyrighted works by Anthropic to determine whether any of the uses qualified as fair uses under the Copyright Act.[ii] The Court ruled:  

1. Anthropic’s use of the works to train its AI models constituted a fair use under the Copyright Act;
2. Anthropic’s digitization of books that it purchased in print form was a fair use because Anthropic replaced the print copies it purchased for its central library with digital copies, without adding new copies of the works, redistributing the existing copies, or creating new works; and
3. It was not a fair use for Anthropic to use pirated copies of copyrighted works to create a permanent and general-purpose library.[iii]

Below is some background on the lawsuit and five takeaways from the 32-page Bartz order on fair use.  

Background of the Lawsuit 

A group of authors brought a class action lawsuit against Anthropic PBC, alleging that Anthropic has built its multibillion-dollar business through the theft of hundreds of thousands of copyrighted works.[iv] Anthropic is the maker of the Claude large language models.[v] The complaint alleges a claim of direct copyright infringement.[vi]  

On March 27, 2025, Anthropic filed a motion for summary judgment on all claims.[vii]  

Additionally, on March 27, 2025, plaintiffs moved for class certification.[viii] On May 23, 2025, the Hon. William Alsup entered an order permitting the parties to attempt to negotiate a class settlement prior to class certification.[ix]  

On June 23, 2025, the Hon. William Alsup issued the order on fair use.[x]  

Five Takeaways from the Order on Fair Use 

1. The Bartz opinion is limited to input-related infringement allegations. 

The Order on Fair Use emphasized repeatedly that the Bartz case does not involve any allegations of AI outputs that infringe on plaintiffs’ works reaching users of Anthropic’s Claude services, and that in the event of such allegations, there would be a different analysis.[xi] The Court explained at one point,  

To repeat and be clear:  Authors do not allege that any LLM output provided to users infringed upon Authors’ works.  Our record shows the opposite.  Users interacted only with the Claude service, which placed additional software between the user and the underlying LLM to ensure that no infringing output ever reached the users.  

* * * 

Here, if the outputs seen by users had been infringing, Authors would have a different case.  And, if the outputs were ever to become infringing, Authors could bring such a case.  But that is not this case.[xii]  

2. LLM training as a fair use was not a close call. 

The Bartz order concluded that three of the four fair use factors supported a finding of fair use for the copies used to train LLMs, and additionally commented, “[t]he technology at issue was among the most transformative many of us will see in our lifetimes.”[xiii] In relation to the first factor, the purpose and character of the use, the Court employed striking language to describe the transformative nature of using the works to train LLMs, including that it was “exceedingly transformative[,]” “transformative – spectacularly so[,]” and “quintessentially transformative.”[xiv]  

3. The use of plaintiffs’ works to train LLMs was distinguished from the Thomson Reuters fair use ruling. 

In February 2025, the Hon. Stephanos Bibas found that Ross Intelligence had infringed 2,243 of Thomson Reuters’ headnotes, and that all of Ross Intelligence’s defenses failed, including fair use, in Thomson Reuters Enter. Centre GmbH v. Ross Intell. Inc.[xv] The Bartz order on fair use was distinguished from the fair use decision in Thomson Reuters, stating, 

* * * [T]he judge there twice emphasized while discussing “purpose and character” of the use that what was trained was “not generative AI (AI that writes new content itself).”  Rather, what was trained — using a proprietary system for finding court opinions in response to a given legal topic — was a competing AI tool for finding court opinions in response to a given legal topic.  That was not transformative.  Thomson Reuters Enter. Centre GmbH v. Ross Intell. Inc., 765 F. Supp. 3d 382, 398 (D. Del. 2025) (Judge Stephanos Bibas), appeal docketed, No. 25-8018 (3d Cir. Apr. 14, 2025). 

A better analogue to our facts would be an AI tool trained — using court opinions, and briefs, law review articles, and the like — to receive legal prompts and respond with fresh legal writing.  And, on facts much like those, a different court came out the other way.  It found fair use.  White v. W. Pub. Corp., 29 F. Supp. 3d 396, 400 (S.D.N.Y. 2014) (Judge Jed Rakoff).  

The latter use stood sufficiently “orthogonal” to anything that any copyright owner rightly could expect to control.  See Warhol, 598 U.S. at 538–40.  It could thus be freed up for the copyist to use, “promot[ing] the progress of science and the arts, without diminishing the incentive to create.”  Id. at 531 (emphasis added); see U.S. CONST. art. I, § 8, cl. 8.        

In short, the purpose and character of using copyrighted works to train LLMs to generate new text was quintessentially transformative.  Like any reader aspiring to be a writer, Anthropic’s LLMs trained upon works not to race ahead and replicate or supplant them — but to turn a hard corner and create something different. * * *[xvi]  

4. The emerging AI licensing market did not help the plaintiffs win on fair use for training LLMs. 

In its analysis of the final fair use factor, the effect of the use upon the market for or value of the copyrighted work, the Court addressed the emerging AI licensing market, stating,

Authors next contend that training LLMs displaced (or will) an emerging market for licensing their works for the narrow purpose of training LLMs (Opp. 21–22).  Anthropic argues that transactional costs would exceed Anthropic’s expected benefit from any such bargain, prompting it to cease dealing with any rightsholders or else to cease developing such technology altogether (Br. 22–23).  Our record could support either account — so this order must assume Authors are correct.  A market could develop (Opp. 19–21 (citing record)).  Even so, such a market for that use is not one the Copyright Act entitles Authors to exploit.    

None of the cases cited by Authors requires a different result.  All contemplated losses of something the Copyright Act properly protected — not the kinds of fair uses for which a copyright owner cannot rightly expect to control. * * *[xvii]  

5. The Court left the door open in relation to any other copies made from Anthropic’s library copies of the books. 

The Court noted that the record was “too poorly developed” to rule in relation to additional copies (other than those used for training) made from the library Anthropic assembled.[xviii] Additionally, the Court stated that Anthropic engineers had made additional copies, that Anthropic had “dodged discovery[,]” and was “not entitled to an order blessing all copying ‘that Anthropic has ever made after obtaining the data,’ * * *.”[xix] The order ended with the following:  

We will have a trial on the pirated copies used to create Anthropic’s central library and the resulting damages, actual or statutory (including for willfulness).  That Anthropic later bought a copy of a book it earlier stole off the internet will not absolve it of liability for the theft but it may affect the extent of statutory damages.  Nothing is foreclosed as to any other copies flowing from library copies for uses other than for training LLMs.[xx]  

Check back next week for the next issue on the Kadrey et al. v. Meta Platforms, Inc. fair use decision. If you haven’t signed up yet, you can have my newsletter delivered to your inbox weekly by signing up here. 

Thanks for being here. 

Jennifer Ballard
Good Journey Consulting

[i] Order on Fair Use, Bartz et al. v. Anthropic PBC, No. 3:24-cv-05417 (N.D. Cal. filed Aug. 19, 2024). 

[ii] Id. at 1. 

[iii] Id. at 9. 

[iv] Complaint at 1, 3, Bartz et al. v. Anthropic PBC, No. 3:24-cv-05417 (N.D. Cal. filed Aug. 19, 2024). 

[v] Id. at 1. 

[vi] Id. at 16-17. 

[vii] Defendant Anthropic PBC’s Notice of Motion and Motion for Summary Judgment at viii, Bartz. 

[viii] Plaintiffs’ Notice of Motion and Motion for Class Certification, Bartz. 

[ix] Permission to Negotiate Class Settlement at 1, Bartz. 

[x] Order on Fair Use, Bartz. 

[xi] Id. at 7, 11-12, 25, 26. 

[xii] Id. at 11-12. 

[xiii] Id. at 30. 

[xiv] Id. at 9, 11, 13.  

[xv] Memorandum Opinion at 2, 5, 14-15, Thomson Reuters Enterprise Centre GMBH et al. v. Ross Intell. Inc., No. 1:20-cv-00613 (D. Del. filed May 6, 2020), (pet. to appeal granted by 3^rd. Cir.). 

[xvi] Order on Fair Use at 13-14, Bartz. 

[xvii] Id. at 28. 

[xviii] Id. at 31. 

[xix] Id. 

[xx] Id. at 31-32.