Training is not free: a wave of lawsuits and the licensing market
„AI companies deliberately trained on pirated sources and must pay; courts increasingly recognize that training data has a market and a price.”
Counter-narrative
Training on lawfully acquired works is transformative fair use, while a publisher-friendly ruling would turn copyright holders into gatekeepers. This position is held by OpenAI, Udio, the US Department of Justice and Wikimedia.
Carriers
New York Times and news publishers, music publishers (UMG, Sony, Warner, Round Hill), author class-action plaintiffs, plaintiff law firms, Copyright Alliance, Hollywood agencies and creator coalitions
Analysis
According to the narrative, the lawsuits are not disputes of principle but about setting a price. This is indicated by the $3,000 per-book payment, Round Hill's claim of more than $1 billion, and the Legal 500 reading that the Third Circuit recognized the licensing market. Unsealed documents strengthen it as evidence of awareness: a Microsoft executive called AI the "greatest work theft in history," and according to the filings, executives knew that piracy was unlawful. For rights holders and plaintiff firms, the stake is bargaining power; for publishers, it is licensing revenue. Its weak points are that Alsup held training on lawfully acquired material to be fair use, and that the Thomson Reuters case narrows to the headnotes of a legal database. The number of lawsuits (more than 110, then 150) does not yet prove infringement in court.
What's at stake: Whether payment is required for training data and who sets the price
Narrative sub-threads
Pirated books and internal documents
According to commentators, Anthropic's court-approved $1.5 billion settlement arose from pirated books, not from training. Judge Alsup ruled that training on legally purchased books constitutes fair use. According to unsealed filings in the authors' lawsuit, OpenAI and Microsoft executives knew that mass pirating of books was illegal.
In this branch, the evidence lies in the manner of the violation and corporate awareness. It is provided by unsealed court documents, executive statements and compensation of about $3,000 per book, not by the fair use doctrine or a sector-wide wave of lawsuits.
▲ at peak · evidence: strongMusic industry lawsuit wave against Suno and Anthropic
Publishers and rights holders are suing Suno, Udio, and Anthropic in a series of cases, with Round Hill claiming over $1 billion and Reservoir approximately $338 million. Meanwhile, Suno is signing licensing agreements, which experts say raises new questions regarding its fair use arguments.
Here, a single sector, the music industry's rights holders, are the carriers, and the mechanism is litigation stacking and parallel licensing. European judicial support comes from a Munich court ruling largely in favor of GEMA.
↗ rising · evidence: moderateNews publishers' fair use front
In the wave of publisher lawsuits surrounding the New York Times case, publishers demand payment for content. The US Department of Justice, by contrast, supports the fair use defense of OpenAI and Microsoft.
The parties are news publishers and the US Department of Justice. The decisive mechanism is the pre-trial fair use ruling and the amicus battle, not judicial recognition of pirate sources or the licensing market.
↗ rising · evidence: moderateJudicial recognition of the training license market
On September 29, the Third Circuit upheld a ruling rejecting the fair use defense in Thomson Reuters v. ROSS. A Legal 500 analysis states the decision recognizes that there is a licensing market for AI training data.
This branch relies on appellate doctrine. The evidence is market harm and legal recognition of the licensing market, not the number of lawsuits or the mode of infringement.
↗ rising · evidence: moderateHungarian cases before the Court of Justice of the EU
In the Hungarian Like Company–Google case (C-250/25), the Court of Justice of the EU is ruling on AI training for the first time, and the preliminary question C-806/26 from the Pécs Court brings another Hungarian case before the court. According to Advocate General Szpunar's opinion, it may be illegal under certain conditions for an AI chatbot to retrieve and reuse online news articles in real time.
Proceeding via EU legal preliminary ruling route, on Hungarian initiative. Focuses on Gemini and protection of press products, not on US fair use lawsuits.
Sub-thread stories
Pirated books and internal documents
Anthropic faces new lawsuit over books used for training as scanning practice comes to lightlawyerThe distinction between fair use and lawful acquisition is the key question in training cases.Bartz case becomes the benchmark for AI copyright: fair use and $1.5 billioncorporate lawyerThe origin of the data source (lawfully acquired or pirated copy) must be documented, because that is where the risk is decided.lawyerIt is advisable to assess the ruling and the settlement separately when advising clients.Anthropic's $1.5 billion settlement: pirated books, not training, triggered the paymentcorporate lawyerFor clients developing models, the provenance of training data should be audited, because infringing acquisition is an independent basis for liability.lawyerClients should be advised on the distinction between fair use and pirated sources.Anthropic settlement: publishers' claims about author payouts are inaccurate, Writer Beware sayslawyerWhen representing authors, the contractual basis for the publisher and author shares should be re-examined.Unsealed briefs: OpenAI and Microsoft executives knew book piracy was unlawful, authors saylawyerThe content of discovery between the parties may change the fair use and intent arguments.Court filings show OpenAI knew training AI on pirated books was illegallawyerIn copyright cases, disclosure of internal communications can be key evidence.Authors' groups urge publishers to give up Anthropic settlement claims on out-of-print bookslawyerSeparate contractual arguments should be prepared for distribution disputes between authors and publishers.21 Korean authors' books eligible for compensation in Anthropic copyright case, including Han KangAnthropic flags insurance gaps and copyright risk after $1.5bn settlementcorporate lawyerLiability insurance for AI developers and users should be checked to confirm whether copyright claims are actually covered.compliance specialistGiven the risk disclosure, a supplier's litigation exposure should be treated as a separate item in vendor risk assessments.
Music industry lawsuit wave against Suno and Anthropic
Munich court rules AI music training and outputs infringe copyright in GEMA v SunolawyerIn cross-border AI training cases, the questions of EU jurisdiction and applicable law gain a new argument.Round Hill files $1 billion copyright lawsuits against Suno and Anthropiccorporate lawyerA licensing risk assessment should be carried out before using AI music and other generative services.lawyerPreparing new types of large-value claims for music rights holders is becoming realistic.Sony and Warner music publishers sue Anthropic over pirated music workscorporate lawyerAI providers must document their training data sources, because the risk of per-work damages is high.Judge allows UMG and Sony to pursue Suno over YouTube stream-ripping and output claimslawyerIn discovery, documents on Suno's training data sources can be requested.Suno licenses and litigates in parallel, straining its fair use argumentlawyerThe licensing agreements can be cited against the other side's litigation arguments, which should be factored into litigation strategy.Udio says AI training is fair use and damages should be as low as $200 per worklawyerBecause the court earlier rejected the expansion of the claims, filing a new complaint may be the route for the plaintiff side.UMG and Sony file second lawsuit against Suno over new modellawyerParallel US and European proceedings call for a joint strategy.UMG and Sony file new suit against Suno; Suno admits scraping YouTube audiolawyerThe relationship between model retraining and earlier infringement (the "poisoned tree" argument) is a new legal question in the cases.Anthropic asks court to rule lyrics training is fair use in dispute with music publisherslawyerThe fair use arguments (harm, purpose of use) and the judge's remarks in the case should be built into strategy in similar disputes.Owners of Elvis Presley catalogue sue Anthropic over 274 songslawyerListing works song by song has become the standard structure of such complaints, which should be considered both when drafting claims and when preparing a defence.Reservoir sues Suno and Anthropic, seeking nearly $338 millionlawyerSimilar legal arguments and evidentiary patterns can be seen across the parallel suits.Eminem's publisher and the Elvis estate both sue AI music platform SunolawyerFor music copyright clients, the voice-imitation issue should be included in the claim alongside the damages demand.corporate lawyerMusic AI providers need to review the legal basis of their training datasets and their licensing agreements.
News publishers' fair use front
US government backs OpenAI and Microsoft in New York Times copyright caselawyerFor clients building a fair use argument, the government's position should be worked into the line of reasoning.corporate lawyerFor AI-developer clients, this signal should be considered in assessing training-data risk, but it does not replace a court ruling.Local newspapers sue OpenAI and Microsoft over use of paywalled articleslawyerCircumvention of a paywall can be a separate basis for a claim alongside the copyright claim.Times, OpenAI and Microsoft ask judge to rule on AI fair use before triallawyerThe fair use argument moves to an early, document-based stage, so evidence strategy must be adjusted accordingly.corporate lawyerThe outcome bears on the risk assessment of content-licensing models tied to AI products.26 local publishers sue OpenAI and Microsoft over copyright infringementlawyerThe number of copyright lawsuits over AI keeps growing, so it is worth tracking the legal grounds of the claims.Justice Department backs OpenAI's fair-use defense in copyright case; news publishers objectlawyerFilings that rely on or attack a fair-use defense must take into account the differing positions of the Copyright Office and the Department of Justice.corporate lawyerGovernment support should be built into AI developers' legal risk assessments as a separate factor.USA Today publisher sues OpenAI for $250 million over copyright infringementlawyerThe fate of §1202(b)(1) claims in the SDNY offers guidance for how to frame complaints.Publishers file amicus brief against OpenAI's fair use defense in copyright caselawyerThe arguments in the fair use debate should be tracked on both sides for use in similar cases.House Democrats criticize Justice Department support for OpenAI in copyright suitlawyerThe weight of the government's brief should be factored into fair use arguments.Wikimedia warns a bad AI copyright ruling could threaten the open weblawyerThe brief offers a new argument for defending transformative use.New York Times and three other US papers sue Brave over AI scraping and paywall bypasscorporate lawyerFor products offering AI search or summaries, companies should check how the system gains access to paywalled content.compliance specialistThe legal risk of AI features that retrieve third-party content needs to be assessed.Hollywood, creators' groups and publishers back New York Times in OpenAI caselawyerCopyright clients should prepare for the arguments in the amicus briefs, which outline the counter-arguments to fair use.corporate lawyerAI developers and users should review the legal basis of their training data while the case remains undecided.
Judicial recognition of the training license market
Appeals court upholds Thomson Reuters win: AI training on its material is not fair uselawyerIn AI copyright cases, the ruling can serve as a starting point for arguments against fair use.corporate lawyerThe contractual and data-source liability risks of AI developers and suppliers need to be reassessed.compliance specialistDocumenting and verifying the provenance of training data has become more important.Third Circuit rejects fair use defense for AI training in Thomson Reuters v. Rosscorporate lawyerThe risk in AI-developer clients' training-data sources and fair use arguments should be reassessed.lawyerThe reasoning in Ross should be analysed for copyright cases, as it may affect the defenses of generative AI defendants.Ninth Circuit weighs AI training and copying as another court rejects fair use in first AI copyright appeallawyerAppellate rulings are shaping what the fair use defense means, which affects clients' litigation strategy.corporate lawyerAn AI vendor's risk depends on whether its training data is legally clean.Thomson Reuters v. ROSS: Third Circuit recognizes licensing market for AI training datalawyerThe existence of a licensing market may strengthen market-harm arguments in copyright cases.Ross Intelligence seeks new hearing in copyright case against Thomson Reuterscorporate lawyerLegal AI vendors still need to watch the lawfulness of their training data.
Hungarian cases before the Court of Justice of the EU
Hungarian local media's lawsuit against Google: EU Court of Justice to rule on AI training for the first timecorporate lawyerFor clients developing or using AI in the EU, the training-data risk assessment should be aligned with the judgment.lawyerIn Hungarian copyright practice, the EU ruling may provide guidance for similar claims.Hungarian court refers Gemini copyright case to EU Court of Justice (Like Company, C-250/25)lawyerIn Hungarian AI copyright cases, it is worth following how the preliminary ruling procedure develops.Google's Gemini copyright case returns to EU top court calendar as third referrallawyerThe third reference will shape the EU interpretation of AI training and the text-and-data-mining exception.Google cases could test EU AI training rules and rights reservationcorporate lawyerChanges to EU rules may affect how training data is sourced.compliance specialistHandling of opt-out signals should be built in now.EU court adviser: Gemini-type chatbots may infringe press publishers' rightslawyerFor publisher clients, claims based on infringement of the related right should be prepared, following the EU court's ruling.corporate lawyerChatbot and AI search providers should review how they retrieve and display press content in real time.Hungarian court makes new EU referral on press publications and LLM chatbotslawyerThe pending case may offer guidance for filings in press copyright and chatbot disputes.
Google search interest
Hungary
| Topic and search term | Weekly | Average |
|---|---|---|
| Training is not free: a wave of lawsuits and the licensing marketAI copyright | 0 | |
| Pirated books and internal documentsAnthropic settlement | 0 | |
| Music industry lawsuit wave against Suno and Anthropicsuno per | 0 | |
| News publishers' fair use frontnew york times openai | 0.1 | |
| Judicial recognition of the training license marketfair use | 1.1 | |
| Hungarian cases before the Court of Justice of the EUpress publishers' right | 0 |
World
| Topic and search term | Weekly | Average |
|---|---|---|
| Training is not free: a wave of lawsuits and the licensing marketai copyright lawsuit | 3.9 | |
| Pirated books and internal documentsanthropic settlement authors | 0.1 | |
| Music industry lawsuit wave against Suno and Anthropicsuno lawsuit | 2.8 | |
| News publishers' fair use frontnew york times openai | 11.1 | |
| Judicial recognition of the training license marketfair use ai | 12.5 | |
| Hungarian cases before the Court of Justice of the EUlike company google | 21.5 |
Google Trends relative search index over the period studied, weekly average. A narrative and its sub-threads appear in a single comparison: 100 is the group's most-searched day, and the figures can only be compared within the group. The search term was chosen by machine analysis; it measures how much the term is searched, not how widespread the narrative is.
Who is affected and how
- corporate lawyer
- When selecting AI suppliers, inquire about the origin of training data and copyright indemnification, as procurement poses the main risk, not training.
- lawyer
- Proving market harm has become the decisive arena, so preparation must involve license market data and experts. The 'market dilution' theory, according to researchers, is not yet supported by evidence.
Hungary
Among the supporting stories for the reviewed period, there is no domestic [HU] story. The narrative has no or minimal domestic trace in the reviewed period.
Weekly trend
Turning points
Anthropic's $1.5 billion settlement: pirated books, not training, triggered the paymentAnthropic's $1.5 billion settlement was paid for pirated books, not for training; this shifted the focus to the legality of acquisition.US government backs OpenAI and Microsoft in New York Times copyright caseThe Department of Justice sided with OpenAI and Microsoft, giving state weight to the counter-narrative against fair use.Appeals court upholds Thomson Reuters win: AI training on its material is not fair useThe Third Circuit upheld Thomson Reuters' victory over ROSS on September 29. Sources say this is the first appellate decision not to consider AI training as fair use.Hollywood, creators' groups and publishers back New York Times in OpenAI caseMore than 50 amicus briefs have been filed in the Times lawsuit, with Hollywood and creator organizations on the Times' side. The debate has widened into an industry front line.
Related narratives in the News magazine
Supporting stories
Anthropic faces new lawsuit over books used for training as scanning practice comes to lightlawyerThe distinction between fair use and lawful acquisition is the key question in training cases.AI copyright suits: Adobe class action, NYT–Microsoft, Nvidia, Redditcorporate lawyerThe litigation risk facing AI developers and their suppliers is also showing up in investor communications.lawyerClass actions are coming to the fore on the plaintiff side, and jurisdictional and claim-narrowing strategies on the defendant side.Round Hill files $1 billion copyright lawsuits against Suno and Anthropiccorporate lawyerA licensing risk assessment should be carried out before using AI music and other generative services.lawyerPreparing new types of large-value claims for music rights holders is becoming realistic.Bartz case becomes the benchmark for AI copyright: fair use and $1.5 billioncorporate lawyerThe origin of the data source (lawfully acquired or pirated copy) must be documented, because that is where the risk is decided.lawyerIt is advisable to assess the ruling and the settlement separately when advising clients.Anthropic's $1.5 billion settlement: pirated books, not training, triggered the paymentcorporate lawyerFor clients developing models, the provenance of training data should be audited, because infringing acquisition is an independent basis for liability.lawyerClients should be advised on the distinction between fair use and pirated sources.Sony and Warner music publishers sue Anthropic over pirated music workscorporate lawyerAI providers must document their training data sources, because the risk of per-work damages is high.WikiHow sues OpenAI for copyright infringement over AI trainingcorporate lawyerFor clients building AI, the legal risk of data collected from public websites should be reassessed.Suno licenses and litigates in parallel, straining its fair use argumentlawyerThe licensing agreements can be cited against the other side's litigation arguments, which should be factored into litigation strategy.US government backs OpenAI and Microsoft in New York Times copyright caselawyerFor clients building a fair use argument, the government's position should be worked into the line of reasoning.corporate lawyerFor AI-developer clients, this signal should be considered in assessing training-data risk, but it does not replace a court ruling.Udio says AI training is fair use and damages should be as low as $200 per worklawyerBecause the court earlier rejected the expansion of the claims, filing a new complaint may be the route for the plaintiff side.More than 110 AI copyright lawsuits filed as AI-generated books flood online marketplacesTimes, OpenAI and Microsoft ask judge to rule on AI fair use before triallawyerThe fair use argument moves to an early, document-based stage, so evidence strategy must be adjusted accordingly.corporate lawyerThe outcome bears on the risk assessment of content-licensing models tied to AI products.Unsealed filings: Microsoft and OpenAI executives knew AI threatened publisherslawyerThe parties' internal communications can become a central tool of discovery, so targeted requests for similar documents are warranted on the publisher side.corporate lawyerAI developers' internal remarks should also be weighed when assessing liability and licensing risk.26 local publishers sue OpenAI and Microsoft over copyright infringementlawyerThe number of copyright lawsuits over AI keeps growing, so it is worth tracking the legal grounds of the claims.Unsealed briefs: OpenAI and Microsoft executives knew book piracy was unlawful, authors saylawyerThe content of discovery between the parties may change the fair use and intent arguments.Justice Department backs OpenAI's fair-use defense in copyright case; news publishers objectlawyerFilings that rely on or attack a fair-use defense must take into account the differing positions of the Copyright Office and the Department of Justice.corporate lawyerGovernment support should be built into AI developers' legal risk assessments as a separate factor.Appeals court upholds Thomson Reuters win: AI training on its material is not fair uselawyerIn AI copyright cases, the ruling can serve as a starting point for arguments against fair use.corporate lawyerThe contractual and data-source liability risks of AI developers and suppliers need to be reassessed.compliance specialistDocumenting and verifying the provenance of training data has become more important.USA Today publisher sues OpenAI for $250 million over copyright infringementlawyerThe fate of §1202(b)(1) claims in the SDNY offers guidance for how to frame complaints.Anthropic asks court to rule lyrics training is fair use in dispute with music publisherslawyerThe fair use arguments (harm, purpose of use) and the judge's remarks in the case should be built into strategy in similar disputes.Hundreds of authors file new copyright claims against Meta over AI traininglawyerThe strategy of mass individual claims is an alternative to class actions, so it means a new client-acquisition and litigation model on the plaintiffs' side.Copyright suits against AI companies in the US reach 150lawyerThe market-effect theory, which is central to fair use outcomes, must be supported with evidence.corporate lawyerCase trackers (class action trackers) help assess risk in dealings with AI vendors.Owners of Elvis Presley catalogue sue Anthropic over 274 songslawyerListing works song by song has become the standard structure of such complaints, which should be considered both when drafting claims and when preparing a defence.Thomson Reuters v. ROSS: Third Circuit recognizes licensing market for AI training datalawyerThe existence of a licensing market may strengthen market-harm arguments in copyright cases.Wikimedia warns a bad AI copyright ruling could threaten the open weblawyerThe brief offers a new argument for defending transformative use.Hollywood, creators' groups and publishers back New York Times in OpenAI caselawyerCopyright clients should prepare for the arguments in the amicus briefs, which outline the counter-arguments to fair use.corporate lawyerAI developers and users should review the legal basis of their training data while the case remains undecided.
Machine analysis based on story titles and summaries, with a verification pass. The source of the facts is the article linked at each story.