A) ABSTRACT / HEADNOTE

Bartz et al. v. Anthropic PBC, No. C 24-05417 WHA concerns a copyright class action against Anthropic PBC arising from alleged copying of books from pirate libraries for artificial intelligence inputs. The order is not a final merits judgment on copyright infringement. It is a Memorandum Opinion on Preliminary Approval of Class Action Settlement. The United States District Court for the Northern District of California granted preliminary approval to the settlement, the plan of distribution, and the plan for notice. The Court found that the settlement was likely to satisfy Federal Rule of Civil Procedure 23(e)(2) at final approval. The proposed class concerned copyright owners of books listed in the Works List from versions of LibGen or PiLiMi downloaded by Anthropic. The settlement offered destruction of covered pirate-library copies, a release limited to past AI input claims, and about $3,000 per work before costs and fees. The opinion is significant for AI copyright litigation, settlement governance, class notice, statutory damages, and digital piracy claims.

Keywords: Copyright class action, artificial intelligence inputs, Anthropic, LibGen, PiLiMi, preliminary approval, Rule 23, settlement notice, statutory damages.

B) CASE DETAILS

i) Judgement Cause Title

Andrea Bartz and Kirk Wallace Johnson, individually, and Andrea Bartz, Inc., Charles Graeber, and MJ + KJ, Inc., individually and as representatives of the class v. Anthropic PBC. The plaintiffs sued Anthropic PBC in a copyright action concerning alleged copying of books. The order specifically addressed preliminary approval of a proposed class settlement. It did not finally adjudicate liability. It did not finally approve the settlement. It memorialised the main points behind preliminary approval. The Court also addressed changes to notice documents, because preliminary approval depended on those changes. The order therefore belongs to the settlement stage. Its legal value lies in its careful treatment of class certification, fairness review, notice, release scope, allocation, and settlement administration in a major AI copyright case.

ii) Case Number

The case number is No. C 24-05417 WHA. The document is Document 437. It was filed in the United States District Court for the Northern District of California. The case is styled as Bartz et al. v. Anthropic PBC. The order records that the plaintiffs brought suit on August 19, 2024. The parties signed a term sheet on August 25, 2025. They signed the proposed settlement agreement on September 5, 2025. The Court held hearings on September 8 and September 25. The second hearing ended in preliminary approval. These dates show a fast but extensive litigation path. The Court noted major discovery before settlement.

iii) Judgement Date

The memorandum opinion is dated October 17, 2025. The fairness hearing was scheduled for noon on April 23, 2026. This timing is important. The order granted preliminary approval only. Final approval remained pending. Class members still had to receive notice. They also had opportunities to object, opt out, submit claims, or contest allocation. The Court repeatedly reserved final assessment until after notice and objections. Thus, the October 17 order should not be cited as final approval. It is a preliminary approval order. It concludes that final approval is likely, not certain. It also directs the parties and appointees to implement the notice and distribution plan carefully.

iv) Court

The Court was the United States District Court for the Northern District of California. The matter was before United States District Judge William Alsup. This is a federal district court order. It is not an appellate judgment. Its persuasive strength comes from its detailed settlement scrutiny. The Court examined Rule 23 standards, class certification, the settlement release, distribution mechanics, notice content, claims administration, and appointment of settlement officials. For Indian readers, this case is useful not as binding law but as comparative guidance. Indian class action mechanisms differ. Yet the order gives insight into judicial control over mass copyright settlements involving AI companies and large groups of authors and publishers.

v) Quorum

The matter was decided by a single district judge, William Alsup, United States District Judge. The rendered image on page 14 shows the judge’s signature and the concluding direction that preliminary approval was granted and notice was directed. The page also records appointments, including JND Legal Administration as settlement administrator, proposed appointment of Attorney Naomi Jane Gray as special master for settlement claims upon satisfactory declaration and order, and continuation of Lieff Cabraser Heimann & Bernstein, LLP and Susman Godfrey LLP as class counsel. This single-judge structure is ordinary for federal district class settlement approval. There is no dissent or concurring opinion.

vi) Author

The author is United States District Judge William Alsup. The opinion is concise but analytically dense. It separates legal standards, preliminary approval, best practicable notice, further notice upon work-specific actions, appointments, and conclusion. The author’s approach is supervisory. He does not merely endorse counsel’s agreement. He tests the settlement against the class action rules. He examines whether the class is likely certifiable for judgment purposes. He evaluates settlement fairness under implied and express factors. He also insists on notice changes. This gives the order a governance character. It shows that AI copyright settlements require careful court management because absent rightsholders may be affected by releases and allocation decisions.

vii) Citation

The citation may be stated as Bartz et al. v. Anthropic PBC, No. C 24-05417 WHA, Document 437, Memorandum Opinion on Preliminary Approval of Class Action Settlement, United States District Court for the Northern District of California, October 17, 2025. The order cites Federal Rule of Civil Procedure 23(e)(1)(B), Rule 23(e)(2), Rule 23(c)(2)(B), 17 U.S.C. §§ 410(c), 412(2), 501(b), 504(c), and 508(a), and several class settlement authorities. It also references Authors Guild v. Google, Inc., 770 F. Supp. 2d 666 (S.D.N.Y. 2011) while comparing settlement value.

viii) Legal Provisions Involved

The central procedural law was Federal Rule of Civil Procedure 23. The Court applied Rule 23(e)(1)(B) for preliminary settlement approval. It also applied Rule 23(e)(2) for likely final approval. It applied Rule 23(c)(2)(B) for best practicable notice. Copyright provisions included 17 U.S.C. § 410(c), concerning registration certificates as evidence, 17 U.S.C. § 412(2), concerning statutory damages timing, 17 U.S.C. § 501(b), concerning infringement actions and notice to interested copyright claimants, 17 U.S.C. § 504(c), concerning statutory damages, and 17 U.S.C. § 508(a), concerning notice to the Copyright Office. Indian comparison may involve Section 14 and Section 51 of the Copyright Act, 1957. However, the uploaded order itself is based on United States law.

ix) Judgments Overruled by the Case

No judgment was overruled. The order did not overrule any authority. It applied class settlement precedents. It relied on Churchill Village, L.L.C. v. General Electric, 361 F.3d 566 (9th Cir. 2004) for implied settlement factors. It referred to Briseño v. Henderson, 998 F.3d 1014 (9th Cir. 2021) and In re Bluetooth Headset Products Liability Litigation, 654 F.3d 935 (9th Cir. 2011) for collusion concerns. It cited Jones v. Flowers, 547 U.S. 220 (2006) and Mullane v. Central Hanover Bank & Trust, 339 U.S. 306 (1950) for notice. It distinguished no major case in a merits sense. The order is administrative and settlement-focused.

x) Related Law Subjects

The case relates to copyright law, class actions, artificial intelligence law, civil procedure, statutory damages, settlement administration, notice, digital piracy, author rights, publisher rights, and technology litigation. It also relates to evidence and claims administration because class members must prove ownership of reproduction rights. The order is important for AI law because the Court expressly stated that “This case concerned AI inputs, not AI outputs.” That distinction is critical. It means the settlement release covered past input-related claims, not output-related claims. It also excluded future conduct after August 25, 2025. This makes the case important for separating training-input copying, pirate-library downloading, future AI conduct, and output-based infringement claims.

xi) Name of Counsels Appearing for Parties

The order does not list all individual advocates in the captioned manner usually seen in merits opinions. It identifies class counsel as Lieff Cabraser Heimann & Bernstein, LLP and Susman Godfrey LLP. It also records that counsel recommended JND Legal Administration after reviewing competing bids from four firms. It states that plaintiffs’ counsel committed to bear administrator costs if the settlement fails. The order does not provide a full list of defence counsel names. Therefore, it would be inaccurate to invent names. The safe statement is that class counsel remained the two named firms. The Court dealt with counsel’s notice revisions and commitments during preliminary approval.

C) INTRODUCTION AND BACKGROUND OF JUDGEMENT

The case arose from alleged copyright violations linked to Anthropic’s downloading of books from pirate libraries. The settlement class covered copyright owners of qualifying books appearing on the Works List. The class definition required ownership of the exclusive right to reproduce copies of any book in the versions of LibGen or PiLiMi downloaded by Anthropic. The work also needed an ISBN or ASIN. It also needed timely registration with the United States Copyright Office. The Court clarified that only works on the Works List were in the class. This was not a cosmetic clarification. It solved notice and release problems. If a work was outside the list, its owner was outside the class for that work. This avoided future ambiguity about released claims.

The settlement followed extensive litigation. The Court noted that discovery included tens of depositions, hundreds of thousands of pages, and terabytes of data. Discovery was close to completion when the term sheet was signed. The parties were therefore reasonably informed. This supported preliminary approval. The Court also referred to its prior rulings. It had ruled that Anthropic’s reproduction of named plaintiffs’ books during LLM training was fair use. It had also ruled that converting purchased print copies into digital copies for a central library, for space-saving and searchability, was fair use. But the Court had ruled that Anthropic’s downloading of millions of books from pirate libraries and assembling them into a central library was not fair use on the summary judgment record. This background shaped settlement value.

The settlement therefore rested on risk. Plaintiffs had a strong case on pirate-library downloading. But victory was not assured. At trial, plaintiffs would need to prove facts to a jury. They would also need to prove willfulness for higher statutory damages. The Court noted that damages could range from $150,000 per work down to $200 per work under 17 U.S.C. § 504(c). A jury award could also be reduced after trial if unsupported. Appeals could follow. Against that uncertainty, the settlement provided prompt closure. It offered injunctive relief and an approximate $3,000 per-work award before costs and fees. The Court described this as the largest copyright class action settlement in history because it resolved nearly half a million works at once.

D) FACTS OF THE CASE

The named plaintiffs included Andrea Bartz, Kirk Wallace Johnson, Andrea Bartz, Inc., Charles Graeber, and MJ + KJ, Inc.. They acted individually and as class representatives. The defendant was Anthropic PBC. The class representatives remained Andrea Bartz, Inc., Charles Graeber, and MJ + KJ Inc. at the end of the order. The class consisted of beneficial or legal copyright owners of reproduction rights in listed books. These books had to be in the relevant LibGen or PiLiMi versions downloaded by Anthropic. They also needed copyright registration within the stated timing. Excluded persons included Anthropic’s directors, officers, and employees, federal agency personnel, and district court personnel. The Works List became central to class identity and settlement administration.

The parties stipulated, for settlement purposes only, that listed works met the works-related criteria. This included timing of registration, publication, and downloading. This mattered because class members then had to prove only ownership of relevant reproduction rights. Authors and publishers could both be class members. The Court noted that no publishers were serving as representatives. It did not treat that as fatal at preliminary approval. It stated that authors appeared to be advocating for all, as anticipated by the class certification order. Still, it recognised a possible problem. The absence of publisher representatives, or representatives for some subset, could surface through claims and objections. This careful observation protects adequacy review. It shows that preliminary approval was conditional and open to later objections.

The release was tailored. Class members would release claims broadly related to past AI inputs. These included past pirating and copying of their works up to the point of AI output. The Court expressly stated that class members did notrelease claims about past AI outputs. They also did not release any claims about future conduct on or after August 25, 2025. In return, Anthropic had to destroy copies of class works derived from LibGen and PiLiMi, unless legal preservation requirements applied. Class members also received the per-work monetary award. This narrow release was a major reason for preliminary approval. It avoided the problem seen in the rejected Authors Guild v. Google settlement, where future claims were released.

E) LEGAL ISSUES RAISED

The first legal issue was whether the Court would likely certify the settlement class for judgment. Under Federal Rule of Civil Procedure 23(e)(1)(B), preliminary approval requires likely ability to approve the proposal finally and certify the class for judgment. The Court found classwide resolution likely proper. It relied on reasons from the class certification order. The clarified class remained substantially the same as certified. The Works List removed uncertainty. It also ensured that excluded works did not create ambiguous future releases. This was important because absent class members must know whether their works and claims are included. A vague class definition could impair due process. The Court found the clarified definition workable enough for preliminary approval.

The second legal issue was whether the settlement was likely fair, reasonable, and adequate under Rule 23(e)(2). The Court considered the Ninth Circuit’s implied factors from Churchill Village, L.L.C. v. General Electric, 361 F.3d 566 (9th Cir. 2004). It examined discovery, litigation risk, settlement value, counsel quality, and likely class reaction. It found discovery extensive. It found plaintiffs had a strong downloading claim, but trial risk remained. It noted the potential range of statutory damages. It also valued prompt closure. The settlement gave injunctive relief and meaningful per-work compensation. Therefore, the Court found the settlement within the range likely to receive final approval. Final approval still required notice and objections.

The third issue was whether the settlement showed signs of collusion. The Court applied concerns from In re Bluetooth Headset Products Liability Litigation, 654 F.3d 935 (9th Cir. 2011) and Briseño v. Henderson, 998 F.3d 1014 (9th Cir. 2021). It found likely absence of collusion. Anthropic would not receive money back unless the settlement failed. The release would also fail if the settlement collapsed. Class counsel’s fee request was capped at 25 percent before costs and expenses. Service fees were also capped and required judicial approval. Counsel committed to receive fees in step with distributions to class members, not merely money entering the fund. The Court also addressed side-deal concerns about rightsholder re-inclusion.

The fourth issue was notice. The Court had to order the best notice practicable under Rule 23(c)(2)(B). Notice had to be reasonably calculated to inform absent class members. The Court relied on Jones v. Flowers, 547 U.S. 220 (2006) and Mullane v. Central Hanover Bank & Trust, 339 U.S. 306 (1950). The notice plan had indirect and direct components. Indirect notice included print media, earned media, social media, web advertisements, and a searchable website. Direct notice used registries, author and publisher group data, interested-party information, commercial locating methods, First Class mail, email, and follow-up procedures. The Court also required improvements in notice language, release disclosure, opt-out instructions, and special master consent.

F) PETITIONER / APPELLANT’S ARGUMENTS

The plaintiffs supported preliminary settlement approval. They relied on the strength of their claims concerning Anthropic’s downloading from pirate libraries. Their position gained force from the Court’s earlier ruling that, on the summary judgment record, Anthropic’s downloading of millions of books from pirate libraries and assembling them into a central library was not fair use. The plaintiffs could argue that the copying was large-scale, unauthorised, and commercially significant. They could also argue that classwide treatment was proper because the Works List and common evidence identified the works. For copyright law, this theory connects strongly to 17 U.S.C. § 501, which enables copyright owners to sue for infringement. It also connects to 17 U.S.C. § 504(c), which provides statutory damages.

The plaintiffs also supported the settlement’s monetary and injunctive benefits. The per-work award of about $3,000 was above the ordinary statutory minimum of $750 and far above the innocent infringement minimum of $200. The Court noted that the amount was four times the ordinary minimum and fifteen times the innocent minimum. The plaintiffs could therefore present the settlement as a practical recovery despite trial risk. The injunctive relief was also important. Anthropic had to destroy covered copies derived from LibGen and PiLiMi unless preservation duties applied. This addressed the specific pirate-library source issue. The plaintiffs’ settlement position was not merely monetary. It sought removal of allegedly unlawful input copies from Anthropic’s retained collections.

The plaintiffs further accepted a tailored release. This was strategically important. The release covered past input-related claims. It did not cover past output claims. It did not cover future conduct after August 25, 2025. This protected class members from overbroad waiver. The Court treated this as a central reason for likely fairness. The plaintiffs also supported a notice and distribution plan designed to locate authors, publishers, and other rightsholders. They accepted mechanisms for opt-outs, opt-backs, claims, competing claims, special master review, and payment allocation. Their position therefore combined legal risk management with practical administration. In a case involving nearly half a million works, perfect individual certainty was impossible. The plaintiffs promoted reasonable and layered notice.

G) RESPONDENT’S ARGUMENTS

Anthropic’s position, as reflected in the order, was that settlement was preferable to continued litigation. The Court recorded that Anthropic had obtained important prior rulings. It had succeeded on fair use regarding reproductions of named plaintiffs’ books in the process of training large language models. It had also succeeded on fair use for one-to-one conversion of purchased print copies into digital copies for its central library, when done for space-saving and searchability. These rulings gave Anthropic litigation strength. At trial, plaintiffs still had to prove facts before a jury. Anthropic could contest willfulness. It could also argue reasonable belief in fair use to reduce damages. It could appeal adverse rulings. These risks supported settlement from Anthropic’s side.

Anthropic also obtained a release of past AI-input-related claims. This gave finality for historical conduct involving pirating and copying up to the point of AI output. However, the release was not unlimited. It did not release output claims. It did not release future claims. Anthropic therefore received meaningful peace, but not total immunity. This limitation made the settlement more acceptable to the Court. Anthropic also avoided trial risk on statutory damages. The Court noted that a plaintiffs’ victory could vary widely, including up to $150,000 per work for willful infringement. Given nearly half a million works, theoretical exposure could be enormous. The settlement allowed Anthropic to resolve historical claims with defined payment, destruction duties, and no admission reflected in this order.

Anthropic also participated in the notice and distribution structure. The Court required clarity that money would not revert to defendant unless the settlement failed. It also required rules against improper side deals. The long-form notice had to warn that rightsholders were prohibited from negotiating any quid pro quo for re-inclusion. Demands for payment in exchange for re-inclusion would be reported to the Court. This protected absent rightsholders and settlement integrity. Anthropic’s settlement posture therefore accepted broad administration, notice, claims review, and judicial oversight. It also accepted destruction of pirate-source copies. These obligations were weighed against the benefit of avoiding a complex trial and appeals.

H) RELATED LEGAL PROVISIONS

Federal Rule of Civil Procedure 23(e)(1)(B) governs preliminary approval. It requires the Court to find that it will likely be able to approve the settlement under Rule 23(e)(2) and certify the class for judgment. Rule 23(e)(2) governs final approval factors. These include adequacy of representation, arm’s-length negotiation, adequacy of relief, and equitable treatment among class members. Rule 23(c)(2)(B) requires the best practicable notice, including individual notice to identifiable members through reasonable effort. These procedural rules were central. The Court did not treat settlement as a private bargain alone. It treated it as a judicially supervised compromise affecting absent persons. That is the core function of class action settlement review.

17 U.S.C. § 504(c) was important because statutory damages framed settlement value. The Court stated that plaintiffs could potentially recover from $150,000 per work down to $200 per work depending on findings and willfulness. The ordinary statutory minimum of $750 was used as a benchmark. The settlement’s approximate $3,000 per-work awardcompared favourably at preliminary approval. 17 U.S.C. § 410(c) and 17 U.S.C. § 412(2) were relevant to registration timing and evidentiary consequences. The class definition required timely registration so works could use registrations as evidence of certain facts and pursue statutory damages. The Court therefore tied class membership to copyright registration rules. This made administration more objective.

17 U.S.C. § 501(b) was relevant to notice. The Court considered whether the section required attaching the complaint to direct notice. It concluded that attaching the complaint was not necessary. It found sufficient a class notice summarising the amended complaint and released claims, while directing recipients to case documents and making copies available upon request. 17 U.S.C. § 508(a) was also relevant. Plaintiffs’ counsel filed materials needed to effect notice to the Copyright Office. For Indian comparison, Section 14 of the Copyright Act, 1957 defines copyright rights, including reproduction. Section 51 defines infringement. Section 55 concerns civil remedies. Section 62 concerns jurisdiction. Indian law does not replicate United States class action settlement procedure. Yet the order is useful for mass-rights settlement design.

I) PRECEDENTS ANALYSED BY COURT IN THIS CASE

Churchill Village, L.L.C. v. General Electric, 361 F.3d 566 (9th Cir. 2004) supplied implied settlement fairness factors. The Court used it to examine discovery, risks, value, counsel’s experience, and likely class reaction. Class Plaintiffs v. City of Seattle, 955 F.2d 1268 (9th Cir. 1992) supported the relevance of informed settlement after meaningful discovery. Briseño v. Henderson, 998 F.3d 1014 (9th Cir. 2021) guided settlement scrutiny and express factors. In re Bluetooth Headset Products Liability Litigation, 654 F.3d 935 (9th Cir. 2011) guided collusion analysis. These cases show that courts must test whether a class settlement is truly fair. They cannot simply defer to counsel. The order applied these principles carefully.

Jones v. Flowers, 547 U.S. 220 (2006) and Mullane v. Central Hanover Bank & Trust, 339 U.S. 306 (1950) were central to notice. They express the due process principle that notice must be reasonably calculated to inform absent persons. The Court used them to assess what a person actually desirous of informing absentees would do. Roes, 1–2 v. SFBSC Management, LLC, 944 F.3d 1035 (9th Cir. 2019) was cited for the idea that low claim submissions may indicate notice failure. Silber v. Mabon, 18 F.3d 1449 (9th Cir. 1994) supported that actual notice to every member is not required. Together, these cases shaped the direct, indirect, and follow-up notice plan.

Authors Guild v. Google, Inc., 770 F. Supp. 2d 666 (S.D.N.Y. 2011) was used comparatively. The Court noted that the Bartz settlement’s per-work amount was far greater than the maximum proposed for books in the rejected Google Books settlement. It also noted that the Google Books settlement was rejected for releasing future claims. This comparison mattered. It showed why the tailored release here was preferable. The Court also cited Drazen v. Pinto, 106 F.4th 1302 (11th Cir. 2024) regarding express Rule 23 factors. These precedents together show that the order is grounded in class settlement law, not merely copyright law. Its most important doctrinal contribution lies in settlement administration for AI copyright claims.

H) JUDGEMENT

a) RATIO DECIDENDI

The ratio decidendi is that the proposed settlement was likely certifiable and likely approvable under Federal Rule of Civil Procedure 23. The Court found that the clarified class remained substantially the same as the previously certified class. The Works List made class scope definite. It also removed ambiguity about release. The parties stipulated, for settlement only, that Works List titles met works-related criteria. Therefore, individual class members needed to prove only ownership of the relevant reproduction rights. This supported manageability. The Court also found that final judgment would likely be appropriate classwide. Potential ownership disputes might arise, but the claims process could address them. The possibility of later ownership disputes did not defeat preliminary approval.

The Court also held that the settlement was within the range likely to receive final approval. It relied on extensive discovery, serious litigation risk, quality counsel, prompt closure, tailored release, injunctive relief, and meaningful per-work recovery. Plaintiffs had strength on downloading claims. Yet trial risk remained. Anthropic had fair use rulings on LLM training reproductions and print-to-digital conversions. Willfulness and damages were uncertain. Appeals could follow. The settlement avoided prolonged complexity. It gave class members about $3,000 per work before costs and fees. It required destruction of LibGen and PiLiMi derived copies subject to legal preservation duties. These benefits justified preliminary approval. The Court also found likely absence of collusion.

The Court’s third ratio concerns notice. It found the notice plan to be the best practicable under the circumstances. The plan combined broad indirect notice and individualized direct notice. It used copyright registries, commercial registries, publisher cooperation, author groups, First Class mail, email, web advertisements, earned media, social media advertisements, and a searchable website. It also included follow-up mechanisms for returned mail, bounced emails, unopened receipts, low claim submissions, opt-outs, opt-backs, claims, and competing claims. The Court required notice content improvements. It insisted on clear release language, opt-out instructions, and special master consent. The notice plan therefore satisfied due process and Rule 23(c)(2)(B) at preliminary stage.

b) OBITER DICTA

The order contains important observations about AI copyright litigation. The most notable is that “This case concerned AI inputs, not AI outputs.” This statement is not merely descriptive. It clarifies the settlement’s legal boundary. It means the case concerned past copying and pirating used as inputs. It did not compromise claims about generated outputs. It also did not release future claims. This observation may influence future AI settlements. Parties must specify whether they are resolving input claims, output claims, future training, future licensing, or all of them. Overbroad releases risk judicial rejection. The rejected Google Books settlement stood as a warning. The Court’s insistence on narrow release protects absent class members from losing unknown future rights.

Another significant observation concerns settlement size. The Court described the settlement as the largest copyright class action settlement in history. This was because nearly 482,460 works were identified, leading to about $3,100 per work before adjustments. The statement reflects the scale of AI-era copyright litigation. AI companies may interact with massive corpora. Traditional one-work infringement suits may be inadequate. Class actions may become a major procedural vehicle. However, class actions also raise ownership, notice, allocation, and representation problems. The Court recognised this. It warned that absence of publisher representatives or certain subsets could become a problem if objections reveal conflicts. This observation is prudent. It keeps the fairness inquiry open until class member responses arrive.

c) GUIDELINES

The order does not issue formal guidelines. Yet practical principles emerge. First, AI copyright settlements must define whether they cover inputs, outputs, or both. Second, future claims should not be casually released. Third, a Works List can make class membership and release scope definite. Fourth, settlement-only stipulations can reduce proof burdens while preserving litigation positions if settlement fails. Fifth, per-work equality may be fair where the infringement theory treats works similarly. Sixth, allocation among co-owners can be handled through claims processes. Seventh, notice must combine registry data, commercial tracing, media, web search tools, direct mail, email, and follow-up procedures. Eighth, fee payment should track class payment to avoid misalignment. Ninth, side deals among rightsholders must be prohibited. Tenth, courts should remain alert to absent publishers, authors, and co-owners.

For Indian legal policy, these principles offer comparative value. India lacks a direct analogue to United States copyright class action settlements. Still, mass copyright disputes may arise through representative suits, public interest claims, assignment disputes, or collective licensing. AI companies may face claims from authors, publishers, photographers, musicians, and film producers. Indian courts and policymakers may need mechanisms for notice, opt-outs, claim verification, and allocation. Section 14 and Section 51 of the Copyright Act, 1957 would frame infringement. Section 55would frame civil remedies. Order I Rule 8 of the Code of Civil Procedure, 1908 may become relevant for representative suits. This order shows how procedural design can determine substantive fairness in mass IP disputes.

d) DISSENTING OPINION

There was no dissenting opinion. The order was issued by one district judge. No concurring opinion exists. The absence of dissent is normal at the district court level. It does not transform preliminary approval into final approval. The Court repeatedly left room for later objections. It stated that class members’ reactions were premature to assess. It promised to consider all objections before final approval. It also recognised possible issues concerning publisher representation and ownership allocation. Therefore, the order’s authority is preliminary. It is strong as a settlement-screening order. It is not a final merits decision. It should not be cited as holding that Anthropic infringed copyright. It should be cited as holding that the settlement was likely fair enough to justify notice and further proceedings.

I) CONCLUSION & COMMENTS

This order is important because it shows how courts may manage AI copyright disputes at scale. The dispute concerned alleged copying of books from pirate libraries. The settlement did not try to solve every AI copyright issue. It focused on past AI inputs. That focus made the release acceptable. The Court’s approach is careful and rights-sensitive. It recognised plaintiffs’ strong case on downloading. It also recognised Anthropic’s litigation defences and prior fair use victories. It understood that statutory damages could be enormous or modest. It therefore treated settlement as a rational compromise. The settlement gave money, destruction of covered copies, and prompt closure. It avoided trial and appeals. Yet it preserved output claims and future claims. That balance is the order’s strongest feature.

The opinion is also a model for notice in complex copyright settlements. Authors and publishers are often difficult to locate. Ownership may be divided. Publishing contracts may allocate proceeds differently. Some works may have multiple rightsholders. Some claims may conflict. The Court responded with layered notice and staged follow-up. It required a searchable Works List. It required direct mail, email, broad advertising, press outreach, and commercial tracing. It required additional notice when opt-outs, opt-backs, or claims identify new rightsholders. It required checks to be sent to identified owners even if some had not filed claims. These procedures reflect due process. They also reflect practical realism. Perfect notice is impossible. Reasonable, persistent, and adaptive notice is required. That is the central procedural lesson.

For Indian scholars, the case is significant in two ways. First, it shows that AI copyright disputes may divide into separate legal categories. Training, pirate downloading, lawful purchase conversion, retention, output generation, and future use are not identical issues. Each needs separate legal analysis. Second, it shows that settlement architecture matters. An overbroad settlement may fail. A narrow settlement may succeed. In India, courts and policymakers may need to develop procedures for mass copyright claims involving AI datasets. They may need rules for notifying authors, publishers, performers, producers, and collecting societies. They may also need mechanisms for claim verification and allocation. Bartz et al. v. Anthropic PBC is therefore not merely an American settlement order. It is a practical blueprint for rights-sensitive AI dispute resolution.

J) REFERENCES

a) Important Cases Referred

1. Bartz et al. v. Anthropic PBC, No. C 24-05417 WHA, Document 437, Memorandum Opinion on Preliminary Approval of Class Action Settlement, United States District Court for the Northern District of California, Oct. 17, 2025.
2. Briseño v. Henderson, 998 F.3d 1014 (9th Cir. 2021).
3. Churchill Village, L.L.C. v. General Electric, 361 F.3d 566 (9th Cir. 2004).
4. Class Plaintiffs v. City of Seattle, 955 F.2d 1268 (9th Cir. 1992).
5. In re Bluetooth Headset Products Liability Litigation, 654 F.3d 935 (9th Cir. 2011).
6. Drazen v. Pinto, 106 F.4th 1302 (11th Cir. 2024).
7. Jones v. Flowers, 547 U.S. 220 (2006).
8. Mullane v. Central Hanover Bank & Trust Co., 339 U.S. 306 (1950).
9. United States v. Rivera-Valdes, No. 21-30177, 2025 WL 267255 (9th Cir. Sept. 18, 2025).
10. Roes, 1–2 v. SFBSC Management, LLC, 944 F.3d 1035 (9th Cir. 2019).
11. Silber v. Mabon, 18 F.3d 1449 (9th Cir. 1994).
12. Authors Guild v. Google, Inc., 770 F. Supp. 2d 666 (S.D.N.Y. 2011).
13. R.G. Anand v. Deluxe Films, (1978) 4 SCC 118.
14. Eastern Book Company v. D.B. Modak, (2008) 1 SCC 1.
15. University of Oxford v. Rameshwari Photocopy Services, 2016 SCC OnLine Del 6229.

b) Important Statutes Referred

1. Federal Rule of Civil Procedure 23(e)(1)(B).
2. Federal Rule of Civil Procedure 23(e)(2).
3. Federal Rule of Civil Procedure 23(c)(2)(B).
4. 17 U.S.C. § 410(c).
5. 17 U.S.C. § 412(2).
6. 17 U.S.C. § 501(b).
7. 17 U.S.C. § 504(c).
8. 17 U.S.C. § 508(a).
9. Treasury Regulation Section 1.468B-1.
10. Copyright Act, 1957, Section 51.
11. Copyright Act, 1957, Section 55.
12. Copyright Act, 1957, Section 62.
13. Code of Civil Procedure, 1908, Order I Rule 8.
14. Berne Convention for the Protection of Literary and Artistic Works.
15. Agreement on Trade-Related Aspects of Intellectual Property Rights.