Advance Local Media LLC et al. v. Cohere Inc., 25-cv-1305 (CM)

A) ABSTRACT / HEADNOTE

Advance Local Media LLC et al. v. Cohere Inc., 25-cv-1305 (CM) concerns a major artificial intelligence copyright and trademark dispute. The plaintiffs are prominent news and digital publishers. They alleged that Cohere Inc. copied copyrighted articles for training and retrieval-based outputs. They also alleged that Cohere’s AI product, Command, generated full copies, substantial excerpts, and substitutive summaries of protected works. The Court, through Judge Colleen McMahon, denied Cohere’s partial motion to dismiss. The ruling held that the publishers had plausibly pleaded direct copyright infringement, secondary copyright infringement, trademark infringement, and false designation of origin. The decision is important because it treats AI-generated summaries, hallucinated articles, retrieval tools, and trademark attribution as legally serious issues. It does not decide final liability. It only holds that the claims deserve discovery. The judgment is significant for Indian legal researchers because it closely resembles future disputes under Section 14, Section 51, and Section 52 of the Copyright Act, 1957, and passing-off principles under Indian trademark law.

Keywords: Copyright infringement, Artificial intelligence, Large language models, Substitutive summaries, Secondary liability, Lanham Act, Trademark confusion, Retrieval Augmented Generation.

B) CASE DETAILS

i) Judgement Cause Title

Advance Local Media LLC, Advance Magazine Publishers Inc. d/b/a Conde Nast, The Atlantic Monthly Group LLC, Forbes Media LLC, Guardian News & Media Limited, Insider, Inc., Los Angeles Times Communications LLC, The McClatchy Company, LLC, Newsday LLC, Plain Dealer Publishing Co., Politico LLC, The Republican Company, Toronto Star Newspapers Limited, and Vox Media LLC v. Cohere Inc. The plaintiffs were described collectively as “Publishers.” The defendant was Cohere Inc. The dispute arose from alleged AI copying, reproduction, display, distribution, and trademark misuse. The Court treated the complaint as one among many lawsuits challenging artificial intelligence companies. These lawsuits question whether copyrighted works may be copied for large language models. The case was not decided after trial. It arose at the pleading stage. Cohere filed a partial motion to dismiss. The Court denied that motion. The judgment therefore preserves the publishers’ claims for discovery. It does not finally decide infringement, damages, fair use, or defences.

ii) Case Number

The case number is 25-cv-1305 (CM). The document was filed as Document 59 in the United States District Court. The Court’s order specifically removed Docket No. 49 from the list of open motions. The procedural posture is essential. The Court was not weighing final evidence. It was deciding whether the complaint contained plausible allegations. That distinction matters deeply. Under Federal Rule of Civil Procedure 12(b)(6), factual allegations are accepted as true. Reasonable inferences favour the non-moving party. This allowed the publishers’ detailed examples to carry weight. It also prevented premature rejection of AI-output infringement theories.

iii) Judgement Date

The judgment is dated November 13, 2025. The date matters because the decision belongs to the fast-developing AI copyright litigation wave. The Court noted that more than fifty lawsuits were already pending. These lawsuits challenge the use of copyrighted works by AI companies. The decision therefore sits inside a larger legal movement. It addresses how courts may evaluate AI model outputs at the pleading stage. It also shows judicial caution. The Court refused to dispose of complex factual issues too early. This is particularly relevant for Indian courts. Indian courts may face similar claims involving news articles, LLM training, and subscription-based content.

iv) Court

The Court was the United States District Court for the Southern District of New York. This forum is highly significant. The Southern District of New York has long handled copyright, media, technology, and publishing disputes. It also has growing experience with AI-related litigation. The Court applied American copyright and trademark principles. Yet its reasoning is comparative useful for India. Indian courts may not follow it as binding precedent. Still, they may find it persuasive. The reasoning on substantial similarity, secondary infringement, and passing off can guide Indian analysis. This is especially true where AI systems reproduce expressive content or falsely attribute outputs to publishers.

v) Quorum

The matter was decided by a single judge. The presiding judge was McMahon, J. The judgment reflects a district court decision. It is not a Supreme Court ruling. It is also not an appellate ruling. Its value lies in persuasive reasoning. The Court examined the pleadings, the complaint, incorporated documents, and alleged examples. It did not conduct a trial. It did not make final factual findings. It treated the allegations as true only for Rule 12(b)(6). This approach is legally orthodox. It protects plaintiffs from dismissal where hidden technical systems require discovery. The Court recognised that AI infringement may occur behind closed digital systems.

vi) Author

The author of the judgment is Judge Colleen McMahon. The order is titled “Decision and Order Denying Defendant’s Partial Motion to Dismiss.” The writing style is direct and issue-focused. The Court first summarised the background. It then stated the pleading standard. It finally addressed direct copyright infringement, secondary copyright infringement, and Lanham Act claims. The judgment is structured around Cohere’s objections. Each objection was rejected. The Court found that the complaint plausibly pleaded infringement theories. This does not mean the publishers will certainly win. It only means their legal theories survive dismissal. Discovery will test evidence, causation, copying, user behaviour, and defences.

vii) Citation

The citation is Advance Local Media LLC et al. v. Cohere Inc., 25-cv-1305 (CM), Document 59, United States District Court, Southern District of New York, November 13, 2025. The judgment cites several American authorities. Important cases include Ashcroft v. Iqbal, 556 U.S. 662 (2009), Bell Atlantic Corp. v. Twombly, 550 U.S. 544 (2007), Feist Publications, Inc. v. Rural Telephone Service Co., 499 U.S. 340 (1991), Nihon Keizai Shimbun, Inc. v. Comline Business Data, Inc., 166 F.3d 65 (2d Cir. 1999), Metro-Goldwyn-Mayer Studios Inc. v. Grokster, Ltd., 545 U.S. 913 (2005), and Dastar Corp. v. Twentieth Century Fox Film Corp., 539 U.S. 23 (2003).

viii) Legal Provisions Involved

The main provisions were 17 U.S.C. §§ 106(1), 106(2), 106(3), 106(5), and 501. These provisions concern reproduction, derivative works, distribution, public display, and infringement. The trademark claims arose under 15 U.S.C. § 1114(1)and 15 U.S.C. § 1125(a)(1)(A). These are Lanham Act provisions. They address trademark infringement and false designation of origin. The procedural rule was Federal Rule of Civil Procedure 12(b)(6). For Indian comparison, the closest provisions are Section 14 and Section 51 of the Copyright Act, 1957. Trademark confusion would resemble Sections 29 and 30 of the Trade Marks Act, 1999.

ix) Judgments Overruled by the Case

No judgment was overruled. The Court did not overturn any prior decision. It distinguished and applied earlier authorities. It rejected Cohere’s reliance on Matthew Bender & Co. v. West Publishing Co., 158 F.3d 693 (2d Cir. 1998) at the pleading stage. It also declined to apply Sony Corp. of America v. Universal City Studios, Inc., 464 U.S. 417 (1984) in the broad manner suggested by Cohere. The Court read Sony through later principles, including Metro-Goldwyn-Mayer Studios Inc. v. Grokster, Ltd., 545 U.S. 913 (2005). It held that the complaint alleged active conduct and intentional design. That made the case different from mere sale of a lawful technology.

x) Related Law Subjects

The case relates to copyright law, trademark law, technology law, artificial intelligence law, media law, unfair competition law, civil procedure, internet law, and digital platform liability. It also has relevance for constitutional free speech debates and international intellectual property law. It intersects with the Berne Convention for the Protection of Literary and Artistic Works and the TRIPS Agreement at a broader level. These instruments protect authors’ rights internationally. They also permit national systems to create limitations and exceptions. The case does not directly apply these instruments. Still, they provide the global legal background. The dispute concerns the balance between innovation and creative labour.

xi) Name of Counsels Appearing for Parties

The provided judgment states that the order was sent “BY ECF TO ALL COUNSEL.” It does not list the individual counsel names in the extracted decision. Therefore, counsel names cannot be accurately supplied from this document. It would be unsafe to invent names. The only reliable statement is that counsel appeared through the federal electronic filing system. The analysis must remain faithful to the judgment. Where the judgment is silent, the researcher should state that silence. This is especially important in legal writing. Accuracy is more important than decorative completeness.

C) INTRODUCTION AND BACKGROUND OF JUDGEMENT

The dispute arose from Cohere’s development and licensing of large language models. Its main product was the Command Family of models. Cohere marketed Command as a “knowledge assistant” suited to business users. The plaintiffs alleged that Command was designed to shortcut research and content analysis. They also alleged that Cohere promoted Command as a tool for receiving the latest news. This factual background mattered. The Court treated news access as central to the business model alleged in the complaint. The plaintiffs claimed that Cohere copied internet text through crawlers and bots. They also alleged use of the Common Crawl dataset. That dataset allegedly contained copyrighted materials and public domain content together. The complaint further alleged that Cohere crawled specific publisher websites. These allegations supported claims of direct copying. They also supported knowledge-based secondary liability. The Court did not decide whether copying actually occurred. It held that the allegations were plausible enough for discovery.

The technical feature at the centre was Retrieval Augmented Generation, or RAG. The judgment explains that LLMs are trained on original datasets. They may lack current information. RAG allegedly permits Command to access external data sources while generating responses. The plaintiffs claimed that Cohere used publisher websites as content sources for RAG. According to the complaint, when RAG was active, Command reproduced copyrighted content. It allegedly delivered full copies, substantial excerpts, and substitutive summaries. The Court considered these allegations legally meaningful. It did not accept Cohere’s framing that summaries were only factual restatements. The complaint pleaded expressive copying. The publishers alleged that Command copied phraseology, organisation, writing style, punctuation, and paragraphs. This distinction between facts and expression became decisive. Under Feist Publications, Inc. v. Rural Telephone Service Co., 499 U.S. 340 (1991), facts are not protected. But original expression in factual writing can be protected.

D) FACTS OF THE CASE

The plaintiffs are major publishers operating news, magazine, and digital platforms. They alleged that their copyrighted works were copied by Cohere. Cohere is a Canadian artificial intelligence company. It develops, operates, and licenses AI models. Its funders included large technology companies. The complaint focused on Cohere’s Command models. Users could access Command through chat interfaces. They could pay based on prompt and output length. They could also deploy their own instances under agreements. Cohere also offered free trials. The plaintiffs alleged that the free trial helped Cohere convert users into paying customers. The Playground interface included an “Under the Hood” feature. That feature allegedly allowed users to view sources used by Command. It could also show full copies of documents used to generate outputs. This was crucial. It strengthened display and reproduction allegations. It also raised concern about paywalled articles.

The publishers alleged two different operational harms. First, with RAG turned on, Command allegedly reproduced actual copyrighted content. This included full verbatim articles, substantial excerpts, and substitutive summaries. The outputs allegedly appeared even when users made ordinary natural-language queries. Users did not always request a specific article. Some prompts asked generally about a subject. Still, Command allegedly delivered protected expression. The complaint contained seventy-five examples. Fifty allegedly included verbatim copying. Twenty-five allegedly involved close paraphrasing and some verbatim copying. One cited example involved a prompt about the unknowability of undecided voters. Command allegedly copied eight of ten paragraphs from a New Yorker article with minor changes. Second, with RAG turned off, Command allegedly hallucinated article text. It falsely attributed fabricated articles to publishers by using their marks. This created trademark and false designation concerns.

E) LEGAL ISSUES RAISED

The first issue was whether the publishers plausibly pleaded direct copyright infringement under 17 U.S.C. §§ 106 and 501. The Court had to decide whether alleged substitutive summaries could be substantially similar to protected expression. Cohere argued that summaries were legally insufficient. It said they only used abstracted facts. The Court rejected dismissal. It held that the complaint alleged more than factual repetition. It alleged expressive copying, close paraphrasing, and near-identical outputs. The issue was not whether every summary infringed. The issue was whether some pleaded examples plausibly did. The Court answered yes. This approach is consistent with Nihon Keizai Shimbun, Inc. v. Comline Business Data, Inc., 166 F.3d 65 (2d Cir. 1999). Quantitative copying must be viewed with qualitative significance. There is no rigid word-count rule.

The second issue was whether the publishers plausibly pleaded secondary copyright infringement. They advanced three theories. These were contributory infringement by material contribution, contributory infringement by inducement, and vicarious infringement. Cohere argued that no underlying direct infringement was adequately alleged. It also argued that the complaint lacked actual knowledge of specific third-party infringement. The Court rejected both arguments. It held that investigator examples may support plausibility. Infringement through AI systems may happen privately. Users’ interactions are not visible to third parties. The Court also applied the Second Circuit standard. Under Arista Records, LLC v. Doe 3, 604 F.3d 110 (2d Cir. 2010), knowledge is objective. Liability may arise when a defendant knows or has reason to know of direct infringement. Specific actual knowledge was not required at this stage.

The third issue concerned trademark infringement and false designation of origin under the Lanham Act. The publishers alleged that Command hallucinated articles while using publisher marks. Cohere argued lack of use in commerce. It also argued lack of likely confusion. Finally, it invoked nominative fair use. The Court rejected dismissal. It held that the complaint plausibly alleged use in commerce. The marks were allegedly displayed to consumers through free trials designed to generate paying customers. The Court also found plausible confusion. The marks were allegedly indistinguishable from the publishers’ marks. Users could believe hallucinated articles were authentic. The Court rejected early application of nominative fair use. False attribution is not merely identification. A trademark cannot be used to pass off fabricated content as another’s work.

F) PETITIONER / APPELLANT’S ARGUMENTS

The counsels for the plaintiffs submitted that Cohere copied their protected works at multiple stages. They alleged copying for training datasets. They also alleged copying through RAG-based retrieval. They stated that Command could provide full articles, large excerpts, and substitutive summaries. These summaries allegedly copied expression, structure, style, and sentence choices. The publishers relied on seventy-five examples. These examples were central to pleading plausibility. They argued that AI outputs were not mere factual digests. They were close substitutes for original articles. This mattered commercially. A user receiving the output may not visit the publisher’s site. The publisher may lose subscription revenue, advertising revenue, licensing value, and reader engagement. The plaintiffs’ theory reflects the copyright principle that facts remain free. But original expression does not. This is close to Indian law under R.G. Anand v. Deluxe Films, (1978) 4 SCC 118, where substantial similarity and overall impression matter.

The plaintiffs also submitted that Cohere knowingly facilitated third-party infringement. They alleged that Cohere marketed Command as a news-access tool. They pointed to promotions describing access to latest news and summaries. They also alleged that Cohere had notice through copyright notices, terms of service, robots.txt instructions, and a cease-and-desist letter. These facts were used to show knowledge and inducement. The plaintiffs further argued that Sony Corp. of America v. Universal City Studios, Inc., 464 U.S. 417 (1984) did not protect Cohere. Their case was not based merely on a tool capable of lawful and unlawful uses. It was based on alleged intentional design and delivery of copied publisher content. This distinction follows Metro-Goldwyn-Mayer Studios Inc. v. Grokster, Ltd., 545 U.S. 913 (2005). Active encouragement may create liability even when a technology has lawful uses.

On trademark, the plaintiffs submitted that Cohere’s hallucinations injured source identification. When Command fabricated articles and placed publisher marks on them, users could believe the publishers authored or approved the text. This was not a harmless label. It created false attribution. It could damage editorial reputation. It could also mislead consumers about source, sponsorship, affiliation, or approval. The publishers argued that Cohere’s use was commercial because Command was part of a monetised service. Even the free trial served customer acquisition. This made the use connected with commerce. The plaintiffs also argued that nominative fair use could not apply. A party may identify a genuine product by its name. But it may not attribute fake goods to the trademark owner. This resembles Indian passing-off doctrine. Under Cadila Health Care Ltd. v. Cadila Pharmaceuticals Ltd., (2001) 5 SCC 73, confusion must be assessed from market realities and consumer perception.

G) RESPONDENT’S ARGUMENTS

The counsel for Cohere submitted that the direct infringement claim should fail insofar as it concerned substitutive summaries. Cohere argued that summaries were not substantially similar as a matter of law. It said the outputs used abstracted facts and new sentences. Since facts are not copyrightable, Cohere argued that summary outputs could not infringe. It relied on the principle in Feist Publications, Inc. v. Rural Telephone Service Co., 499 U.S. 340 (1991). It also relied on Nihon Keizai Shimbun, Inc. v. Comline Business Data, Inc., 166 F.3d 65 (2d Cir. 1999) to argue that copying limited portions may not constitute infringement. The Court rejected this narrow view. It held that the complaint alleged copying beyond facts. It also noted that Nihon did not create a fixed twenty-percent rule. Qualitative value remains important.

Cohere also submitted that secondary infringement claims were defective. It argued that the publishers had not adequately pleaded direct infringement by ordinary users. It said investigator-generated outputs did not show typical use. It relied on Matthew Bender & Co. v. West Publishing Co., 158 F.3d 693 (2d Cir. 1998). The Court found that case distinguishable. It was decided after discovery. This case was at the pleading stage. Cohere further argued that actual knowledge of specific infringement was required. The Court rejected that standard for this Circuit. Cohere also argued against inducement. It said marketing a product as a latest-news tool was not inducement. The Court disagreed at the pleading stage. The allegations included promotional conduct, system design, and delivery of copied works. That was enough for plausibility.

On the Lanham Act claims, Cohere submitted that outputs were not used in commerce. It argued that users purchased API access, not specific outputs. It also argued that one-off responses were not shared publicly. Cohere further contended that the complaint lacked real-world confusion. It sought application of the Polaroid Corp. v. Polarad Electronics Corp., 287 F.2d 492 (2d Cir. 1961) factors. Cohere also invoked nominative fair use. It argued that marks merely identified requested content. Finally, Cohere relied on Dastar Corp. v. Twentieth Century Fox Film Corp., 539 U.S. 23 (2003) to resist trademark claims over communicative products. The Court rejected dismissal. It held that false attribution of hallucinated articles could constitute classic passing off. It also held that Dastar did not bar passing-off claims.

H) RELATED LEGAL PROVISIONS

The copyright provisions were 17 U.S.C. § 106(1), 17 U.S.C. § 106(2), 17 U.S.C. § 106(3), 17 U.S.C. § 106(5), and 17 U.S.C. § 501. These provisions protect reproduction, derivative works, distribution, public display, and infringement remedies. The plaintiffs alleged that Cohere reproduced copyrighted articles, distributed outputs, displayed copies, and created derivative or substitutive forms. Indian law contains a comparable structure. Section 14 of the Copyright Act, 1957 gives exclusive rights. Section 51 defines infringement. Section 52 provides exceptions. In an Indian AI dispute, the key questions may include reproduction during scraping, storage, training, retrieval, and output generation. Courts would also examine fair dealing. Indian fair dealing is narrower than American fair use. Therefore, the Indian analysis may differ sharply from American doctrine.

The trademark provisions were 15 U.S.C. § 1114(1) and 15 U.S.C. § 1125(a)(1)(A). These provisions protect against unauthorised use of marks and false designation of origin. The Lanham Act also defines “use in commerce” under 15 U.S.C. § 1127. The Court examined whether marks were displayed to consumers in connection with commercial transactions. Indian equivalents include Section 29 of the Trade Marks Act, 1999 for infringement. Section 30 addresses limits on effect of registered marks. Passing off remains preserved under Section 27(2). If an Indian AI tool generated fabricated articles under a publisher’s mark, both infringement and passing off may arise. Reputation-based harm would be serious. The doctrine in Laxmikant V. Patel v. Chetanbhai Shah, (2002) 3 SCC 65 would support protection against deceptive trade representation.

The procedural rule was Federal Rule of Civil Procedure 12(b)(6). The standard resembles Indian rejection of plaint principles only in limited ways. Under American law, the complaint must state a plausible claim. The Court relied on Ashcroft v. Iqbal, 556 U.S. 662 (2009) and Bell Atlantic Corp. v. Twombly, 550 U.S. 544 (2007). The Court accepted pleaded facts as true. It drew reasonable inferences for the publishers. It asked whether discovery could reveal supporting evidence. In India, Order VII Rule 11 of the Code of Civil Procedure, 1908 allows rejection of plaint in specific circumstances. Indian courts also avoid deciding disputed facts prematurely. Thus, the reasoning has comparative value. Complex AI system behaviour requires technical discovery. Early dismissal may be inappropriate where facts are hidden inside proprietary systems.

I) PRECEDENTS ANALYSED BY COURT IN THIS CASE

Ashcroft v. Iqbal, 556 U.S. 662 (2009) and Bell Atlantic Corp. v. Twombly, 550 U.S. 544 (2007) governed pleading sufficiency. The Court used them to test plausibility. A claim need not prove liability at the pleading stage. It must contain enough facts to raise a reasonable expectation of discovery. Feist Publications, Inc. v. Rural Telephone Service Co., 499 U.S. 340 (1991) separated facts from expression. The Court applied that separation to AI summaries. Tufenkian Import/Export Ventures, Inc. v. Einstein Moomjy, Inc., 338 F.3d 127 (2d Cir. 2003) explained actual copying and substantial similarity. Boisson v. Banian, Ltd., 273 F.3d 262 (2d Cir. 2001) supported the more discerning ordinary observer test. Together, these cases framed whether Command copied protected expression, not merely facts.

Nihon Keizai Shimbun, Inc. v. Comline Business Data, Inc., 166 F.3d 65 (2d Cir. 1999) was central. It involved translated and summarised news materials. The Court used it carefully. Cohere wanted a quantitative rule. The Court rejected that. Nihon did not hold that copying twenty percent can never infringe. It held that qualitative nature matters. Ringgold v. Black Entertainment Television, Inc., 126 F.3d 70 (2d Cir. 1997) also supported qualitative and quantitative assessment. Matthew Bender & Co. v. West Publishing Co., 158 F.3d 693 (2d Cir. 1998) was distinguished. That case arose after discovery. This case arose at the pleading stage. The Court also cited Arista Records LLC v. Usenet.com, Inc., 633 F. Supp. 2d 124 (S.D.N.Y. 2009) for investigator-based proof.

For secondary liability, the Court discussed Arista Records, LLC v. Doe 3, 604 F.3d 110 (2d Cir. 2010). It stated the objective knowledge standard. A defendant may be liable where it knows or has reason to know. The Court also relied on The New York Times Co. v. Microsoft Corp., 777 F. Supp. 3d 283 (S.D.N.Y. 2025). That case rejected an actual-knowledge-only standard in similar AI litigation. Capitol Records, LLC v. ReDigi Inc., 934 F. Supp. 2d 640 (S.D.N.Y. 2013), aff’d, 910 F.3d 649 (2d Cir. 2018) supported the business-model inference. Sony Corp. of America v. Universal City Studios, Inc., 464 U.S. 417 (1984) was limited. Metro-Goldwyn-Mayer Studios Inc. v. Grokster, Ltd., 545 U.S. 913 (2005) supported liability for active inducement.

For trademark law, the Court cited Camelot SL LLC v. ThreeSixty Brands Group LLC, 632 F. Supp. 3d 471 (S.D.N.Y. 2022). It stated the basic Lanham Act elements. Rescuecom Corp. v. Google Inc., 562 F.3d 123 (2d Cir. 2009) addressed use in commerce. Kelly-Brown v. Winfrey, 717 F.3d 295 (2d Cir. 2013) supported commercial display analysis. Polaroid Corp. v. Polarad Electronics Corp., 287 F.2d 492 (2d Cir. 1961) supplied confusion factors. But the Court noted that identical marks may not require mechanical factor analysis. Tiffany (NJ) Inc. v. eBay Inc., 600 F.3d 93 (2d Cir. 2010)explained nominative use. Dastar Corp. v. Twentieth Century Fox Film Corp., 539 U.S. 23 (2003) did not bar the pleaded passing-off theory.

H) JUDGEMENT

a) RATIO DECIDENDI

The ratio is that the publishers plausibly pleaded actionable copyright and trademark claims. The Court held that AI-generated substitutive summaries cannot be dismissed merely because they are called summaries. The true question is substantial similarity. If the output copies protected expression, organisation, phrasing, or style, infringement may be plausible. The Court stressed that facts are not protected. But expression is protected. It found the complaint pleaded copying that was both quantitatively and qualitatively significant. The seventy-five examples mattered. The alleged New Yorker example was especially powerful. It showed near-identical copying of several paragraphs. Therefore, whether the summaries infringe is a factual issue. It cannot be decided merely on Cohere’s characterisation. This reasoning is important. It prevents AI defendants from escaping scrutiny through labels like summary, paraphrase, or abstraction. The law looks at substance over terminology.

The Court’s second ratio concerns secondary infringement. It held that underlying direct infringement was plausibly pleaded. Investigator-generated examples could support the claim. The Court recognised that AI infringement may be hidden from third parties. Users interact privately. Outputs are not always publicly visible. This makes discovery important. The Court also held that the publishers plausibly pleaded Cohere’s knowledge. The Second Circuit does not require actual knowledge of specific acts at this stage. Objective knowledge is enough. The publishers alleged copyright notices, terms of service, robots.txt restrictions, and a cease-and-desist letter. They also alleged that Cohere designed Command to deliver copied publisher works. These facts created an inference that Cohere knew or had reason to know. The Court also found inducement plausible. Marketing Command as a tool for accessing latest news supported active encouragement.

The Court’s third ratio concerns trademark and false designation. It held that Cohere’s alleged use of publisher marks could be use in commerce. The free trial was not merely private experimentation. It served customer acquisition. Command’s outputs were part of a commercial service. The Court also held that likely confusion was plausible. The marks were allegedly indistinguishable. The hallucinated articles allegedly appeared under publisher identities. Users could believe that fabricated articles were real publisher content. The Court further rejected early nominative fair use. Nominative use protects truthful identification. It does not protect false attribution. The Court treated the allegation as classic passing off. This ratio may shape future AI hallucination cases. It recognises that AI hallucinations can create trademark harm. Fabricated content under a real mark can injure goodwill, trust, and source integrity.

b) OBITER DICTA

The Court made several observations beyond the narrow holding. It observed that this case forms part of more than fifty lawsuits challenging AI companies. This statement gives institutional context. It shows judicial awareness of a wider transformation in copyright litigation. The Court also observed that deciding substantial similarity through simple word counts is improper. This is not merely procedural. It is a substantive warning. AI outputs may copy value without copying massive quantities. Conversely, small copying may be important when qualitative elements are taken. The Court’s discussion of hidden infringement is also important. It recognised that AI user interactions occur behind closed doors. This supports a practical evidentiary approach. In Indian law, such reasoning may support discovery, interrogatories, and technical audits. Courts should not demand impossible public proof where systems are proprietary.

Another important observation concerns Dastar Corp. v. Twentieth Century Fox Film Corp., 539 U.S. 23 (2003). Cohere tried to rely on communicative-products reasoning. The Court rejected an overbroad reading. It clarified that Dastarprevents copyright-like claims disguised as trademark claims in some reverse passing-off situations. It does not immunise false attribution. The Court also expressed doubt about nominative fair use on these facts. It did not finally decide the defence. It only held that the complaint alleged facts capable of defeating it. This observation matters for AI outputs. A system that fabricates text and labels it with a real publisher’s mark does more than identify content. It may misrepresent origin. That misrepresentation can damage both consumers and publishers. The legal wrong is not only copying. It is also deception about source.

c) GUIDELINES

The judgment does not issue formal guidelines. Still, several practical principles emerge from its reasoning. First, AI summaries must be assessed for protected expression. Courts should not assume summaries are lawful. Second, factual content remains free. But expressive selection, arrangement, phrasing, and style may be protected. Third, AI-output infringement may be pleaded through investigator examples. Plaintiffs need not produce public user examples at the initial stage. Fourth, knowledge may be inferred from notices, robots.txt restrictions, cease-and-desist letters, design choices, and commercial marketing. Fifth, marketing an AI system as a latest-news tool may support inducement. Sixth, hallucinated content under real marks may create trademark confusion. Seventh, nominative fair use cannot protect false attribution. Eighth, dismissal is improper where discovery may reveal technical evidence. These principles are not binding guidelines. They are doctrinal signals. Indian courts may adapt them cautiously under Indian copyright and trademark statutes.

d) DISSENTING OPINION

There was no dissenting opinion. The matter was decided by a single district judge. Therefore, no separate judicial opinion exists. There is also no concurring opinion. The judgment speaks through one judicial voice. The absence of dissent is procedurally expected. A district judge decides motions individually. The lack of dissent does not make the decision final on merits. It only means the partial motion to dismiss was denied. Cohere may still contest copying, substantial similarity, causation, knowledge, fair use, damages, and trademark confusion later. The ruling preserves the claims. It does not predetermine the final result. This distinction is vital for legal research. A pleading decision has persuasive force. But it does not establish final infringement. It establishes that the pleaded case is legally and factually plausible enough to proceed.

I) CONCLUSION & COMMENTS

The judgment is important because it refuses to reduce AI copyright disputes into simple slogans. It does not say that all training is unlawful. It does not say that every AI summary infringes. It also does not accept that summaries are automatically lawful. The Court carefully asks whether protected expression was copied. This is the correct analytical path. Copyright does not protect facts, ideas, systems, or abstract knowledge. It protects original expression. News reporting contains facts. But it also contains expressive choices. Those choices include structure, emphasis, arrangement, tone, and language. Where an AI system reproduces these choices, infringement may become plausible. The Court’s refusal to dismiss protects the evidentiary process. It allows technical discovery into model design, retrieval systems, source copying, logs, user prompts, output patterns, and internal knowledge. This is necessary in AI litigation. Without discovery, plaintiffs may never access the system evidence needed to prove their case.

From an Indian perspective, the case is highly instructive. Indian copyright law gives authors and owners exclusive rights over reproduction, communication, adaptation, and issuing copies. If an Indian AI system copied protected publisher works and generated substitute outputs, courts would examine Section 14 and Section 51 of the Copyright Act, 1957. The fair dealing defence under Section 52 would be central. However, Indian fair dealing is purpose-specific. It is not identical to American fair use. Training and commercial retrieval may face narrower protection in India. Indian courts would also examine moral rights under Section 57 where attribution or distortion harms reputation. For trademarks, false attribution of hallucinated articles could trigger Section 29 of the Trade Marks Act, 1999 and passing off. Indian cases like R.G. Anand v. Deluxe Films, (1978) 4 SCC 118 and Cadila Health Care Ltd. v. Cadila Pharmaceuticals Ltd., (2001) 5 SCC 73 would help frame similarity and confusion. The judgment therefore offers persuasive comparative reasoning for Indian AI disputes.

The most significant feature is the Court’s treatment of hallucination. AI hallucination is often discussed as a technical weakness. This judgment treats it as a legal risk. When hallucinated content carries a real publisher’s trademark, the harm is not imaginary. It may mislead readers. It may degrade brand trust. It may falsely connect publishers to content they never authored. That injury resembles classic passing off. The Court’s Chanel analogy, adopted from the publishers’ argument, is powerful. A trader cannot label non-Chanel handbags as Chanel. Likewise, an AI system cannot safely label fabricated articles as works of real publishers. This insight will matter globally. AI systems increasingly generate plausible but false content. Where that content uses protected marks, law must protect both consumers and brand owners. The ruling does not punish innovation. It demands legal accountability. That balance is likely to shape future AI governance.

J) REFERENCES

a) Important Cases Referred

  1. Advance Local Media LLC et al. v. Cohere Inc., 25-cv-1305 (CM), Document 59, U.S. District Court for the Southern District of New York, Nov. 13, 2025.
  2. Ashcroft v. Iqbal, 556 U.S. 662 (2009).
  3. Bell Atlantic Corp. v. Twombly, 550 U.S. 544 (2007).
  4. Feist Publications, Inc. v. Rural Telephone Service Co., 499 U.S. 340 (1991).
  5. Tufenkian Import/Export Ventures, Inc. v. Einstein Moomjy, Inc., 338 F.3d 127 (2d Cir. 2003).
  6. Nihon Keizai Shimbun, Inc. v. Comline Business Data, Inc., 166 F.3d 65 (2d Cir. 1999)
  7. Boisson v. Banian, Ltd., 273 F.3d 262 (2d Cir. 2001).
  8. Ringgold v. Black Entertainment Television, Inc., 126 F.3d 70 (2d Cir. 1997).
  9. Matthew Bender & Co. v. West Publishing Co., 158 F.3d 693 (2d Cir. 1998).
  10. Arista Records, LLC v. Doe 3, 604 F.3d 110 (2d Cir. 2010).
  11. Arista Records LLC v. Usenet.com, Inc., 633 F. Supp. 2d 124 (S.D.N.Y. 2009)
  12. UMG Recordings, Inc. v. RCN Telecom Services, LLC, 2020 WL 5204067 (D.N.J. Aug. 31, 2020).
  13. Warner Bros. Records, Inc. v. Payne, 2006 WL 2844415 (W.D. Tex. July 17, 2006).
  14. The New York Times Co. v. Microsoft Corp., 777 F. Supp. 3d 283 (S.D.N.Y. 2025).
  15. Capitol Records, LLC v. ReDigi Inc., 934 F. Supp. 2d 640 (S.D.N.Y. 2013), aff’d, 910 F.3d 649 (2d Cir. 2018).
  16. Sony Corp. of America v. Universal City Studios, Inc., 464 U.S. 417 (1984).
  17. Metro-Goldwyn-Mayer Studios Inc. v. Grokster, Ltd., 545 U.S. 913 (2005).
  18. Andersen v. Stability AI Ltd., 744 F. Supp. 3d 956 (N.D. Cal. 2024).
  19. Camelot SL LLC v. ThreeSixty Brands Group LLC, 632 F. Supp. 3d 471 (S.D.N.Y. 2022).
  20. Lopez v. Nike, Inc., 2021 WL 128574 (S.D.N.Y. Jan. 14, 2021).
  21. 1-800 Contacts, Inc. v. JAND, Inc., 119 F.4th 234 (2d Cir. 2024).
  22. Rescuecom Corp. v. Google Inc., 562 F.3d 123 (2d Cir. 2009).
  23. Kelly-Brown v. Winfrey, 717 F.3d 295 (2d Cir. 2013).
  24. 1-800 Contacts, Inc. v. WhenU.com, Inc., 414 F.3d 400 (2d Cir. 2005).
  25. Polaroid Corp. v. Polarad Electronics Corp., 287 F.2d 492 (2d Cir. 1961).
  26. Ryan v. Volpone Stamp Co., 107 F. Supp. 2d 369 (S.D.N.Y. 2000).
  27. C=Holdings B.V. v. Asiarim Corp., 992 F. Supp. 2d 223 (S.D.N.Y. 2013).
  28. Hectronic GmbH v. Hectronic USA Corp., 2020 WL 6947684 (S.D.N.Y. Nov. 24, 2020).
  29. Municipal Credit Union v. Queens Auto Mall, Inc., 126 F. Supp. 3d 290 (E.D.N.Y. 2015).
  30. Savin Corp. v. Savin Group, 391 F.3d 439 (2d Cir. 2004).
  31. Star Industries, Inc. v. Bacardi & Co., 412 F.3d 373 (2d Cir. 2005).
  32. Pulse Creations, Inc. v. Vesture Group, Inc., 154 F. Supp. 3d 48 (S.D.N.Y. 2015).
  33. Dow Jones & Co., Inc. v. International Securities Exchange, Inc., 451 F.3d 295 (2d Cir. 2006).
  34. Tiffany (NJ) Inc. v. eBay Inc., 600 F.3d 93 (2d Cir. 2010).
  35. Dastar Corp. v. Twentieth Century Fox Film Corp., 539 U.S. 23 (2003).
  36. Beastie Boys v. Monster Energy Co., 66 F. Supp. 3d 424 (S.D.N.Y. 2014).
  37. Williams v. Cavalli, 2015 WL 1247065 (C.D. Cal. Feb. 12, 2015).
  38. Grand v. Schwarz, 2016 WL 2733133 (S.D.N.Y. May 10, 2016).
  39. International Council of Shopping Centers, Inc. v. Info Quarter, LLC, 2018 WL 4284279 (S.D.N.Y. Sept. 7, 2018).
  40. R.G. Anand v. Deluxe Films, (1978) 4 SCC 118.
  41. Cadila Health Care Ltd. v. Cadila Pharmaceuticals Ltd., (2001) 5 SCC 73.
  42. Laxmikant V. Patel v. Chetanbhai Shah, (2002) 3 SCC 65.

b) Important Statutes Referred

  1. 17 U.S.C. § 106(
  2. 17 U.S.C. § 106(2).
  3. 17 U.S.C. § 106(3).
  4. 17 U.S.C. § 106(5).
  5. 17 U.S.C. § 501.
  6. 15 U.S.C. § 1114(1).
  7. 15 U.S.C. § 1125(a)(1)(A).
  8. 15 U.S.C. § 1127.
  9. Federal Rule of Civil Procedure 12(b)(6).
  10. Copyright Act, 1957, Section 14.
  11. Copyright Act, 1957, Section 51.
  12. Copyright Act, 1957, Section 52.
  13. Copyright Act, 1957, Section 57.
  14. Trade Marks Act, 1999, Section 27(2).
  15. Trade Marks Act, 1999, Section 29.
  16. Trade Marks Act, 1999, Section 30.
  17. Code of Civil Procedure, 1908, Order VII Rule 11.
  18. Berne Convention for the Protection of Literary and Artistic Works.
  19. Agreement on Trade-Related Aspects of Intellectual Property Rights.

Share this :
Facebook
Twitter
LinkedIn
WhatsApp
Picture of Jaya Dubey
Jaya Dubey