GPT-4o: Martinez-Conde v. Apple is a meticulously constructed and potentially explosive copyright infringement case that places Apple’s AI ambitions under judicial scrutiny.

Legal Analysis of Martinez-Conde v. Apple – Copyright Infringement in AI Training

by ChatGPT-4o

Introduction

In Martinez-Conde v. Apple, filed on October 9, 2025, two prominent neuroscientists and authors—Susana Martinez-Conde and Stephen L. Macknik—initiated a class-action lawsuit against Apple Inc. for direct copyright infringement. Represented by the Joseph Saveri Law Firm, they allege that Apple used their copyrighted work, along with the works of other authors, to train its generative AI system “Apple Intelligence” without permission or compensation. The case focuses on the unauthorized use of works from the Books3 dataset—a widely condemned shadow library—by Apple in its foundational and OpenELM AI models. This lawsuit joins a broader wave of legal actions by authors against tech companies in the generative AI space.

Nature of the Grievances

The plaintiffs assert that:

1. Apple unlawfully reproduced copyrighted works for AI training without consent.

2. Their book, Sleights of Mind, was found in the Books3 dataset, which Apple used to train OpenELM and Foundation Language Models.

3. Apple’s conduct impaired the market for their work by creating tools that substitute for human-authored content.

4. Apple failed to license the works despite being aware of industry norms and legal risks, including existing licensing practices and known copyright issues with datasets like Books3.

They claim not just infringement, but willful infringement, and demand statutory damages, actual damages, disgorgement of profits, and destruction of all infringing models under 17 U.S.C. §503(b). They also seek class certification to represent potentially thousands of similarly affected authors.

Quality of the Evidence and Argumentation

The plaintiffs provide a detailed, well-researched narrative that includes:

• Identification of the infringing models: Apple’s OpenELM and Foundation Language Models.

• Clear documentation of Apple’s public statements and papers that confirm the use of Books3 via RedPajama and The Pile—both of which are known to include pirated materials.

• Precise metadata: They cite ISBNs, GitHub model cards, Hugging Face documentation, and Apple’s own AI papers to triangulate Apple’s use of unauthorized datasets.

• Strong chain-of-custody logic: If Apple used Books3 (either directly or through subsets like RedPajama), and if Sleights of Mind is verifiably part of Books3, then Apple must have made unauthorized copies of their work.

The complaint also contextualizes Apple’s actions within broader industry behavior, drawing analogies with other AI firms’ training practices. They stress Apple’s sophisticated filtering and processing steps (e.g., Applebot scraping, tokenization, classifier training), arguing that these constitute additional derivative uses beyond simple copying.

Furthermore, the plaintiffs argue that Apple’s failure to seek licenses, even while licensing other datasets (like Shutterstock or news archives), shows intentional evasion. This weakens any claim of good-faith fair use or accidental inclusion.

Legal Context and Precedents

The lawsuit joins a growing field of litigation at the intersection of copyright and AI. Notably:

• Silverman v. OpenAI and Andersen v. Stability AI test similar questions around AI training on copyrighted text and art.

• Getty Images v. Stability AI involves AI training on visual content and has seen the court reject early motions to dismiss, suggesting momentum for rightsholders.

• The Bartz v. Anthropic case (cited in this complaint) is especially important. Judge Alsup’s decision affirms that copying works from pirate sources like Books3 “is infringement, full stop,” regardless of downstream transformation.

Courts have increasingly signaled skepticism about blanket “fair use” defenses in training contexts involving unauthorized, full-text ingestion of copyrighted material.

Strengths of Plaintiffs’ Case

1. Strong evidentiary trail linking Apple to pirated datasets.

2. Registered copyrights for their works—giving access to statutory damages and the presumption of ownership.

3. Class certification strategy: Inclusion of all authors whose registered works were ingested into Apple’s models could create massive financial and reputational pressure.

4. Growing body of case law sympathetic to creators and skeptical of opaque AI training practices.

5. Detailed description of harm, including market dilution, loss of licensing revenue, and substitution effects.

Apple’s Likely Defenses

Apple may argue:

1. Fair use, citing transformation and lack of direct substitution.

2. Lack of substantial similarity or verbatim copying in outputs.

3. That Books3 was “publicly available” and training was non-commercial or research-focused.

4. That training data was filtered or deduplicated, mitigating the claim of “copying in entirety.”

5. That any restitution or liability lies upstream, with those who compiled and distributed Books3 (e.g., Hugging Face, EleutherAI).

However, courts are becoming wary of these arguments, especially where pirated content is used knowingly, and the defendant is a multitrillion-dollar commercial actor integrating the models into core products.

Predicted Outcome

Chances of Apple winning outright (e.g., on a motion to dismiss): Low to moderate
The detailed tracing of datasets and reliance on Books3, already deemed infringing in other cases, is a major liability. Even if Apple argues that they used “deduplicated” or derivative datasets, this may not absolve them, especially if they retained or benefited from the pirated source material.

Chances of a settlement: High
Apple is unlikely to want discovery into its AI training data pipelines, and the risk of model destruction (as demanded) adds pressure. Settlements could involve:

• Monetary compensation.

• Public commitments to licensing agreements.

• Technical adjustments to future model training.

Chances of plaintiffs winning class certification: Moderate to high
The class is well-defined (registered copyright owners whose books are in Books3), and the legal theory is consistent across members. If certified, Apple faces the risk of statutory damages multiplied by thousands of works, even without needing to prove economic harm in every case.

Broader Implications

This case may become a landmark copyright precedent in AI litigation, especially if Apple is forced to:

• Disclose training datasets in full.

• Destroy infringing models or retrain them from scratch.

• Pay large sums to a collective licensing body.

It also underscores the need for a clean and auditable licensing regime for AI training data, much like the music industry adopted after the Napster era.

Conclusion

Martinez-Conde v. Apple is a meticulously constructed and potentially explosive copyright infringement case that places Apple’s AI ambitions under judicial scrutiny. Given the strength of evidence, the precedent of similar cases, and the reputational risks at stake, the likely outcome is either a plaintiff-friendly settlement or a partial win for the class, with monetary and injunctive relief. Apple’s broad use of datasets like Books3—already a red flag in the courts—may undermine any claim to fair use or innocence, making this case a cautionary tale for all AI developers relying on unlicensed data to train models.