Since our last overview in July 2026, four developments in particular are worth noting:

• On 19 August 2026, German children’s and young-adult publisher Carlsen Verlag, author Marc-Uwe Kling and illustrator Astrid Henn filed an action against OpenAI Ireland Ltd. before the Regional Court of Munich I in Germany. Carlsen is supported by the German Bonnier publishing group; Bonnier is not identified as a claimant in Carlsen’s press release.

- The final approval of the USD 1.5 billion settlement in Bartz v. Anthropic was entered on 20 July 2026. Distribution remains subject to implementation and other post-judgment steps.

- On 24 July 2026, India’s Delhi High Court refused ANI Media’s application for interim relief against OpenAI. At the preliminary stage, it considered OpenAI’s use of ANI material for training capable of falling within India’s fair-dealing exception. The merits proceedings continue.

- The European Commission’s enforcement powers concerning general-purpose AI models became applicable on 2 August 2026. Provider obligations generally applied from 2 August 2025; certain models placed on the market before that date benefit from a transition period until 2 August 2027.

Training data and reproduction

Claimants argue that downloading, storing, processing and incorporating protected works into training datasets can amount to copyright-relevant reproduction. The alleged acquisition of works from shadow libraries such as LibGen and Anna’s Archive is a distinct factual issue in several US cases.

Whether training on lawfully acquired works is permitted is assessed in the United States through fair use. European cases operate under different statutory frameworks, including text-and-data-mining provisions and copyright-reservation rules.

Memorisation and output

Several claimants contend that protected material persists in model parameters and can be retrieved through ordinary prompts in verbatim or recognisably similar form. The Regional Court of Munich I treated particular ChatGPT reproductions of song lyrics as copyright-relevant in the GEMA case.

That ruling does not automatically decide book, illustration or other output cases. Similarity, access and legal responsibility must be assessed separately for each work, model and output.

Market effects, licensing and attribution

Publishers and media groups also allege that AI-generated summaries and substitute content can reduce purchases, referrals, subscriptions or advertising revenue. Some claimants argue that unlicensed use bypasses existing or emerging commercial licensing markets. Others challenge outputs falsely attributed to authors, publishers or media brands. Depending on the jurisdiction and facts, such claims may implicate copyright, trademark, personality or unfair-competition law.

Carlsen Verlag, Marc-Uwe Kling and Astrid Henn v. OpenAI Ireland Ltd.

Pending — Germany, August 2026

According to Carlsen’s press release, the German children’s and young-adult publisher Carlsen Verlag, author Marc-Uwe Kling and illustrator Astrid Henn filed suit against OpenAI Ireland Ltd. before the Regional Court of Munich I on 19 August 2026. OpenAI Ireland operates ChatGPT in Europe. Carlsen is supported by the German Bonnier publishing group.

The dispute concerns the German-language children’s-book series Das NEINhorn and its sequels. The claimants allege that simple prompts cause ChatGPT to generate stories containing essential creative elements of the protected works and illustrations that are barely distinguishable from the originals. They further allege that ChatGPT proactively suggests additional material involving characters and settings not mentioned in the prompt.

Carlsen also says the system offered print-ready material including cover design, an imprint page, an ISBN and a publisher logo. Bonnier CEO Christian Schumacher-Gebler stated in the press release that one generated imprint page named Kling for text, Henn for illustrations and Carlsen Verlag GmbH for copyright.

The claimants infer from the alleged outputs that works from the series may have been used without authorisation to train OpenAI’s language and image models and may be memorised in those models. These are the claimants’ allegations and legal assessment; they have not been established by a court.

Carlsen says it seeks injunctive relief, disclosure concerning possible misconduct and damages. The press release alone does not establish whether distinct trademark or unfair-competition claims form part of the action. The alleged use of an invented ISBN and a publisher logo may raise additional legal questions, but those claims should not be assumed without the pleadings.

Penguin Random House Verlagsgruppe v. OpenAI

Pending — Germany, March 2026

German publishing group Penguin Random House Verlagsgruppe filed an action against OpenAI Ireland Ltd. before the Regional Court of Munich I on 27 March 2026. The case concerns the German children’s-book series Der kleine Drache Kokosnuss by German author and illustrator Ingo Siegner.

The publisher alleges that simple prompts cause ChatGPT to reproduce protected text and illustrations in recognisable form. It also alleges that the system proposes a print-ready manuscript with cover art, promotional copy and instructions for self-publishing. These are allegations made by the publisher, not findings of the court.

The case raises questions not only about alleged training and memorised outputs, but also about the responsibility an AI provider may bear for particular outputs and system-suggested follow-on uses. Whether German law would characterise that responsibility as direct liability, participation, interference liability or another doctrine remains open.

The corporate context has drawn attention: German media group Bertelsmann announced a strategic collaboration with OpenAI in January 2025, including the use of ChatGPT Enterprise in parts of its businesses.

Hachette, Cengage, Elsevier, Scott Turow and S.C.R.I.B.E. v. Google

Pending — United States, July 2026

US publishers Hachette Book Group, Cengage Learning and Elsevier, US author Scott Turow and S.C.R.I.B.E. filed a proposed class action against Google in the US District Court for the Southern District of New York on 10 July 2026: Hachette Book Group, Inc. et al. v. Google LLC, No. 1:26-cv-05870.

The complaint alleges that Google copied millions of copyrighted books and journal articles to train Gemini. It includes claims for direct copyright infringement, contributory infringement and alleged removal or alteration of copyright-management information under the DMCA. The plaintiffs seek, among other remedies, damages and injunctive relief.

The plaintiffs argue that agreements governing access to works through Google Books, Google Play Books or Google Scholar did not authorise AI training. They further contend that a generative model capable of producing competing text differs materially from a search index that directs users to an underlying book. These allegations have not been adjudicated.

Elsevier, Cengage, Hachette, Macmillan, McGraw Hill and Scott Turow v. Meta Platforms and Mark Zuckerberg

Pending — United States, May 2026

US and international publishers Elsevier, Cengage Learning, Hachette Book Group, Macmillan Publishers and McGraw Hill, together with US author Scott Turow, filed a proposed class action against Meta Platforms and Mark Zuckerberg in the US District Court for the Southern District of New York on 5 May 2026: Elsevier Inc. et al. v. Meta Platforms, Inc. et al., No. 1:26-cv-03689.

The plaintiffs allege that Meta acquired millions of copyrighted books and journal articles from piracy sources, including LibGen and Anna’s Archive, and used them in connection with training Llama models. They also allege claims involving copyright-management information. Zuckerberg is named as a defendant; the complaint alleges that he supported the conduct at issue.

No court has ruled on the merits. Class certification had not been decided as of the stated date.

SNE, SGDL and SNAC v. Meta

Pending — France, 2025

French publishers’ association Syndicat national de l’édition (SNE) and French authors’ organisations SGDL and SNAC announced proceedings against Meta in early 2025. Their public statements concern alleged unauthorised use of protected works for generative AI, including alleged reliance on the Books2 and Books3 datasets, and claims framed under French copyright law and parasitisme économique.

The organisations’ public material describes the proceedings as ongoing. The precise claims and procedural posture should be verified against the Paris court file before publication.

Publishers v. Anna’s Archive

Default judgment entered — United States, May 2026

US publisher Apress Media and other publishers sued Anna’s Archive and unnamed defendants in the US District Court for the Southern District of New York on 6 March 2026: Apress Media, LLC et al. v. Anna’s Archive, No. 1:26-cv-01850. Anna’s Archive is not an AI provider; it is a shadow-library and piracy platform.

The publishers alleged unauthorised distribution of copyrighted books and other works, including the alleged provision of high-quality material for AI training. On 19 May 2026, Judge Jed Rakoff entered a default judgment after the defendants failed to appear. The judgment awarded USD 150,000 in statutory damages for each of 130 works, totalling USD 19.5 million, and imposed broad injunctive and site-blocking measures. Practical enforcement of monetary and technical relief remains a separate question.

Bartz v. Anthropic

Settlement finally approved — distribution pending

US authors Andrea Bartz, Charles Graeber and Kirk Wallace Johnson filed Bartz v. Anthropic in the US District Court for the Northern District of California in 2024. They alleged that Anthropic used books to train Claude, including works obtained from shadow libraries.

In 2025, Judge William Alsup distinguished between training on lawfully acquired books and questions concerning pirated copies. He held that training on lawfully acquired books was fair use under the facts before him. The settlement subsequently resolved claims involving the relevant class of pirated works.

The USD 1.5 billion settlement received final approval from Judge Araceli Martínez-Olguín on 20 July 2026. The court reduced requested attorneys’ fees from USD 187.5 million to approximately USD 101.6 million. The settlement covers roughly 482,000 works, yielding a mathematical average of slightly more than USD 3,100 per work; individual awards may differ.

The settlement does not establish a general minimum or maximum benchmark for damages or settlements in other AI copyright disputes. Separate opt-out cases illustrate that some rights holders did not consider the settlement terms adequate, but their claims remain distinct.

The New York Times and other media organisations v. OpenAI and Microsoft

Pending — United States, discovery

US newspaper publisher The New York Times sued OpenAI and Microsoft in the US District Court for the Southern District of New York in late December 2023. Related cases have been centralised in In re OpenAI, Inc. Copyright Infringement Litigation, MDL No. 1:25-md-03143.

The proceedings involve claims brought by, among others, the New York Daily News, other Alden and MediaNews titles, the US nonprofit Center for Investigative Reporting, The Intercept and Ziff Davis, as well as author plaintiffs.

Key issues include the alleged use of journalism and books for training, alleged acquisition of works from shadow libraries, allegedly infringing or substitutive outputs, and the relevance of outputs to fair-use market-effects analysis.

The court ordered OpenAI to produce a sample of 20 million anonymised ChatGPT conversations. On 9 July 2026, the New York Times, the New York Daily News and other media organisations sought sanctions, alleging deficiencies in OpenAI’s discovery conduct. OpenAI disputes those allegations. No trial date had been set as of 20 August 2026.

Advance Local Media, Condé Nast and others v. Cohere

Pending — United States, partial motion to dismiss denied

Fourteen North American and international news and magazine publishers, including Advance Local Media, Condé Nast, The Atlantic, Forbes Media, The Guardian, Business Insider, the Los Angeles Times, McClatchy, Newsday, Politico, the Toronto Star and Vox Media, sued Canadian AI developer Cohere in the US District Court for the Southern District of New York on 13 February 2025, No. 1:25-cv-01305.

The plaintiffs allege verbatim reproduction, substantial excerpts and substitutive summaries through Cohere’s Command models, as well as claims concerning paywall circumvention, trademarks and false attribution.

On 13 November 2025, Judge Colleen McMahon denied Cohere’s partial motion to dismiss. The ruling permits the disputed claims to proceed; it does not establish that Cohere infringed. The case remains relevant because it allows allegations concerning allegedly substitutive summaries and brand-related claims to be tested in discovery.

Encyclopaedia Britannica and Merriam-Webster v. OpenAI

Pending — United States, March 2026

US reference publishers Encyclopaedia Britannica and Merriam-Webster sued OpenAI in the US District Court for the Southern District of New York on 13 March 2026, No. 1:26-cv-02097. The complaint alleges that OpenAI copied close to 100,000 protected reference works for training and retrieval-related functions, and that ChatGPT can reproduce or closely paraphrase material from those sources.

The plaintiffs also allege harm to website traffic, subscriptions and advertising revenue, and bring trademark-based claims concerning invented material allegedly presented under their names. These claims remain unadjudicated.

Britannica and Merriam-Webster had previously sued US AI-search company Perplexity in the same court on 10 September 2025, No. 1:25-cv-07546. That action remains pending.

Yomiuri Shimbun, Nikkei and Asahi Shimbun v. Perplexity

Pending — Japan, August 2025

Japanese news publisher Yomiuri Shimbun brought proceedings against Perplexity in Tokyo in August 2025. It alleges that Perplexity used and reproduced approximately 120,000 articles without permission and seeks damages reported at approximately JPY 2.17 billion.

On 26 August 2025, Japanese publishers Nikkei and Asahi Shimbun jointly sued Perplexity in the Tokyo District Court. Press reports state that each seeks JPY 2.2 billion.

The cases concern alleged unauthorised acquisition and reproduction of news content, possible interference with publishers’ rights, and claimed economic effects on traffic and advertising. Exact statutory claims should be taken from the complaints rather than inferred from media reports.

ANI Media v. OpenAI

Pending — India, interim injunction refused

Indian news agency ANI Media sued OpenAI in India. On 24 July 2026, Justice Amit Bansal of the Delhi High Court refused ANI’s application for an interim injunction. In a 135-page order, the court took the preliminary view that OpenAI’s storage and use of ANI material for training could fall within the fair-dealing exception in section 52(1)(a) of India’s Copyright Act.

The order does not finally determine all claims. The merits case continues, and allegations that ChatGPT generated false reports attributed to ANI require separate assessment. The ruling should not be treated as directly comparable with or binding on the German cases: it concerns a different legal regime, factual record and interim-relief application.

Antitrust and competition proceedings concerning AI summaries

Chegg and Penske Media v. Google

Pending — United States, motions to dismiss unresolved

US education-technology company Chegg sued Google and Alphabet in the US District Court for the District of Columbia, No. 1:25-cv-00543. US media company Penske Media filed a separate action on 12 September 2025, No. 1:25-cv-03192.

The plaintiffs allege that Google uses market power in search to incorporate publisher and media content into AI Overviews and related products without adequate compensation. They further allege that direct AI answers can reduce referral traffic, subscriptions, advertising and other revenue.

Google moved to dismiss both cases. The motions remained unresolved as of 20 August 2026. Chegg and Penske requested oral argument and contend that their claims differ materially from those in earlier litigation brought by Helena World Chronicle and Emmerich Newspapers, which was dismissed in March 2026.

Formal investigation open — European Union

On 9 December 2025, the European Commission opened formal antitrust proceedings to examine whether Google’s use of web-publisher content and YouTube material for AI purposes may breach EU competition law. The investigation concerns AI Overviews and AI Mode, possible unfair conditions for publishers, opt-out mechanisms and potential disadvantage to rival AI developers.

On 10 February 2026, the European Publishers Council filed a further complaint. It called for effective publisher control over AI uses, transparency concerning traffic and revenue effects, and a licensing and remuneration framework. The Commission has not made a final infringement finding.

GEMA v. OpenAI

First-instance judgment; not final — Germany

On 11 November 2025, the Regional Court of Munich I largely upheld German collecting society GEMA’s claims against OpenAI concerning nine German song lyrics, including claims for injunctive relief, information and damages, No. 42 O 14139/24.

The decision is important because it treated particular ChatGPT outputs as copyright-relevant and placed alleged memorisation at the centre of the analysis. It is not a general decision on the permissibility of generative-AI training or on all forms of book and image output. The judgment is not final.

Detailed propositions concerning the InfoSoc Directive, sections 16, 44b and 57 of the German Copyright Act, or separate personality-rights claims should be stated only after checking the full reasons for judgment and the appellate record.

GEMA v. Suno

First-instance judgment; finality to be confirmed — Germany

On 31 July 2026, the Regional Court of Munich I largely upheld GEMA’s claims against US AI music generator Suno concerning six musical works, No. 42 O 763/25. This is another first-instance decision by the specialised Munich chamber in litigation involving an AI provider. Specific holdings on training, storage and outputs should be reported only from the full judgment.

Kadrey v. Meta and Bartz v. Anthropic

The California cases demonstrate the importance of distinguishing among the use of works for training, the provenance of copies and alleged market effects. In Kadrey v. Meta, the court held that the training use at issue was fair use. In Bartz, the court distinguished lawfully acquired books from works obtained through piracy sources.

Neither decision creates a universal rule that all training on lawfully accessible works is permitted or that every use of unlawfully obtained material automatically resolves all copyright claims. The legal analysis remains fact- and claim-specific.

EU AI Act

Provider obligations for general-purpose AI models generally applied from 2 August 2025. They include, among other matters:

- a policy for compliance with EU copyright law;

- measures addressing enforceable rights reservations;

- technical documentation; and

- publication of a sufficiently detailed summary of the content used for training.

From 2 August 2026, the Commission’s enforcement powers in this area became applicable. General-purpose AI models placed on the market before 2 August 2025 generally have until 2 August 2027 to comply.

The public training-content summary is not a complete disclosure of a model’s training corpus. It is designed to provide structured information about types and sources of training material rather than individual rights holders with access to every underlying file.

- The two German book cases: Whether the Munich court finds particular text and image outputs sufficiently similar to protected works, and how it treats the alleged initiative of the system.

- Provider responsibility: Whether and on what basis a provider can be responsible for allegedly infringing outputs or system-suggested follow-on uses.

- The US publisher cases against Meta and Google: The treatment of dismissal motions, class certification and evidence about training-data provenance.

- Discovery in the OpenAI MDL: The 20-million-conversation sample may be relevant to output patterns and the fair-use market-effects inquiry.

- Implementation of the EU AI Act: Training-content summaries and copyright policies may give rights holders useful leads, but they do not provide full access to training corpora.

- Divergent legal systems: The cases in Germany, India, the United States, France and Japan arise under different statutes and procedural postures. Their outcomes should therefore be compared cautiously.

This compilation was produced using the dpr archive and with the help of AI (Claude)