When the Dominoes Fall: What the Anthropic Settlement and Apple’s New Lawsuit Mean for AI Copyright

The $1.5 billion Anthropic settlement may be the largest copyright recovery in U.S. history, but it didn’t close the book on AI and copyright. The release is deliberately narrow — past conduct only, inputs not outputs, and just the works on the official list. That means the real fight ahead is not about the past, but about the future: how to assign value between creators and those who use AI to develop new products.

The Core Issue: IP Infringement in Generative AI

Generative AI systems are trained on massive datasets, often scraped without consent. That reality has sparked lawsuits from authors, coders, musicians, and visual artists — all claiming infringement of their intellectual property.

For AI developers and companies deploying these tools, the exposure is real: liability may flow not just from how the model was trained, but from what it produces and how those outputs are used.

The current wave of lawsuits mainly targets frontier AI companies—the firms building general-purpose foundation models like OpenAI, Anthropic, Meta, and Microsoft. But the legal precedents set in these cases will extend to all AI developers. Cases such as Thomson Reuters v. Ross Intelligence—though not involving generative AI—illustrate that exposure to copyright claims is not limited to the frontier players; downstream and domain-specific AI companies also face significant risk.

What to Watch:

Unlicensed training data vs. lawfully sourced content
Internal experimentation vs. public distribution of outputs
Legal ambiguity around what counts as “transformative” use

This tension defines the landscape: creators demand protection for their works, while AI adopters seek clarity on where the legal red lines truly lie.

What Legal Teams Should Do Now

Legal professionals advising businesses cannot wait for courts to provide definitive answers. Practical steps are needed today:

Audit Your AI Supply Chain
- Know your models, sourcing methods, and whether vendor indemnities are meaningful. Courts are already scrutinizing acquisition methods, and under the EU AI Act, providers must now publish summaries of their training datasets. That transparency obligation is a lever — ask vendors to align to it even outside Europe.
- Importantly, audit outputs. If your AI systems generate content, run controlled tests to see if outputs risk reproducing copyrighted material or confusing consumers. Establish review protocols, train teams on what’s permissible, and create fallback plans for API or dataset reliance. Litigation is showing that outputs will be scrutinized as much as inputs.
Review Vendor Agreements
Require transparency about data provenance. Push vendors to disclose datasets, training methods, and known risks. Where possible, hardwire disclosure and remediation obligations into contracts.
Establish Internal Content Policies
Create workflows to flag risky AI-generated outputs. Build safeguards to prevent models from reproducing substantial chunks of copyrighted works.
Embed Compliance Into Development
Make compliance part of the engineering workflow. Require legal documentation at each stage: data acquisition, preprocessing, fine-tuning, and deployment. Don’t let product teams operate in a legal vacuum.
Negotiate Upstream Deals → Don’t wait for lawsuits. Anticipate the issues and pay it forward: start negotiating licensing arrangements with publishers, rights holders, and content creators now. That way, you control the narrative — instead of being dragged into a fight over whether “publicly available” really means “lawfully used.”

Why Act Now? The Dominoes Are Already Falling

These steps aren’t abstract. The lawsuits driving these risks are already underway. Each case shows how courts are drawing lines — and where they are leaving gaps. Together, they create the chain reaction that explains why proactive steps are essential.

Here’s how the first four dominoes have tipped:

The First Domino: Anthropic’s $1.5 Billion Settlement

On September 5, 2025, the parties in Bartz v. Anthropic filed a motion for preliminary approval of a landmark settlement. Anthropic agreed to pay at least $1.5 billion to authors whose works were scraped from shadow libraries and used to train its AI models. At an estimated $3,000 per work, it may be the largest copyright recovery in U.S. history.

But what really matters is what the deal didn’t cover. The release is narrow in three ways:

Only past conduct (through August 25, 2025) is released.
Inputs only (training data), not outputs.
Only works on the official list — nothing else.

That means the core questions about AI outputs, market harm, and licensing remain wide open.

Update – Sept. 8 Hearing:
That “landmark settlement” is not yet a done deal. At a Sept. 8 hearing, Judge William Alsup sharply criticized the agreement as “full of pitfalls” and postponed preliminary approval. He ordered the parties to deliver a final Works List by Sept. 15 and clarifications on claims procedures by Sept. 22, with a new hearing set for Sept. 25. Alsup’s concerns centered on ensuring the list of ~465,000 pirated books does not expand, that authors are fully informed of their rights, and that industry groups are not exerting undue influence behind the scenes. His warning was blunt: “We’ll see if I can hold my nose and approve it.”

Update (Sept. 25, 2025): Anthropic agreed to a record-setting $1.5 billion class action settlement with authors over the use of pirated books to train its Claude models. Judge William Alsup granted preliminary approval, calling it the largest copyright recovery in U.S. history. The deal covers past conduct only, requires Anthropic to destroy infringing copies, and pays roughly $3,000 per book for 482,460 titles. While the settlement sets a powerful benchmark for valuing training data, it reflects litigation risk rather than a true arm’s length license price — leaving open the larger questions about outputs, future markets, and what fair compensation should look like.

The Other Shoe: Anthropic’s Music Copyright Battle

The author settlement doesn’t end Anthropic’s copyright exposure. In Concord Music Group, Inc. v. Anthropic PBC (N.D. Cal.), major publishers including Universal, Concord, and ABKCO allege that Anthropic copied and trained on pirated song lyrics, stripped copyright management info, and allowed Claude to reproduce those lyrics in outputs.

Plaintiffs seek damages and an injunction that could force Anthropic to implement stricter guardrails or even destroy models trained on infringing lyrics.

Key deadlines stretch well into 2026, with dispositive motion hearings set for June. This case squarely raises the unresolved question of output liability — the very gap left open by the author settlement.

The Second Domino: Eleven Labs Settles Early

Just days before the Anthropic settlement filing, Vacker v. Eleven Labs became the first of nearly fifty AI copyright lawsuits to settle. Filed in the District of Delaware, the case involved voice actors and authors who alleged misuse of their voices and works.

The settlement became a bellwether, showing that plaintiffs are widening their toolkit beyond copyright to include right of publicity and DMCA claims, and that some AI firms may prefer confidential settlements to precedent-setting losses.

The Third Domino: Meta’s Bellwether Win

Meta’s case, Kadrey v. Meta, ended with summary judgment in its favor. Judge Vince Chhabria didn’t bless Meta’s training practices as lawful; rather, he ruled that plaintiffs failed to prove market harm. But in doing so, he signaled the path forward: market dilution claims could succeed if future plaintiffs build the record.

That makes Kadrey less a clean victory for Meta than a bellwether for the next wave of litigation — showing which arguments courts will dismiss, and which they may one day accept.

The Fourth Domino: Apple Enters the Book Fight

On the same day Anthropic’s settlement was filed, authors Grady Hendrix and Jennifer Roberson filed a proposed class action against Apple in the Northern District of California. The complaint alleges that Apple trained its OpenELM and Apple Intelligence models on Books3, a dataset of ~196,000 pirated books.

Core Allegations

Unauthorized copying: Apple allegedly used plaintiffs’ works without consent, credit, or compensation.
Shadow libraries + scraping: Training allegedly relied on Books3/Bibliotik and Applebot scraping, not just public-domain materials.
Market harm: Plaintiffs argue Apple deprived authors of licensing fees in a training data market already valued in the billions — and diluted sales by spawning “sham books” and AI-generated summaries.
Retention: Apple is accused of maintaining a private library of pirated data for future training.
Relief sought: Damages, injunction, and even destruction of models trained on infringing data under § 503(b).

Why Apple’s Case Is Different

Apple isn’t just accused of training proprietary systems. The case also targets OpenELM, its open-source release. When a company publishes open-source weights, it distributes trained parameters that others can run, fine-tune, and redeploy. If infringing works are baked in, the risk ripples outward — multiplying potential claims across the ecosystem.

Strategically, Apple is placing its bets on two tracks of market harm:

The AI licensing market theory — the idea that authors are entitled to licensing fees whenever their works are used in training. Both Judge Alsup in Bartz v. Anthropic and Judge Chhabria in Kadrey v. Meta rejected this theory, finding it too speculative and outside the scope of the Copyright Act.
The market dilution theory — the claim that AI-generated works will flood the marketplace, competing with originals and thinning royalty pools. Judge Alsup dismissed the dilution argument with a sharp analogy: authors’ claims were “no different than if they complained that training schoolchildren to write well would result in an explosion of competing works” — the kind of creative displacement the Copyright Act does not protect against. Judge Chhabria also rejected dilution for lack of evidence — but he emphasized that with a stronger factual record, it could eventually tip the balance against fair use.

In short, Apple isn’t advancing shiny new theories. It’s reviving arguments that other judges have already heard — betting that stronger evidence, plus the procedural weight of a class action, might finally make them stick.

From Inputs to Outputs: The Next Battleground

So far, the spotlight has been on inputs — the datasets scraped, copied, or purchased to train AI models. But as Judge Alsup noted in Bartz v. Anthropic, the case might have looked very different if plaintiffs had alleged infringing outputs. That question — what models actually produce — is now front and center.

Several cases are already testing this terrain:

New York Times v. Microsoft & OpenAI (MDL): The Times alleges ChatGPT can reproduce near-verbatim passages from paywalled articles and mimic its distinctive phrasing. Consolidated in April 2025, the multidistrict litigation is forcing OpenAI to preserve output logs — even deleted user conversations — as potential evidence.
Disney v. Midjourney (June 2025): Disney and Universal allege Midjourney generates unauthorized images of their high-profile animated characters.
Warner Bros. Discovery v. Midjourney (Sept 2025): Plaintiffs Warner Bros. Entertainment Inc., DC Comics, Turner Entertainment Co., Hanna-Barbera Productions, Inc., and The Cartoon Network, Inc. (collectively “Warner Bros. Discovery”) filed suit for direct and secondary copyright infringement under the Copyright Act (17 U.S.C. § 101 et seq.). The case is grounded in U.S. copyright law’s most fundamental protections: only Warner Bros. Discovery has the right to reproduce, prepare derivative works, distribute, publicly display, and perform images and videos featuring characters like Superman, Batman, Wonder Woman, Bugs Bunny, and Scooby-Doo. The complaint argues that these exclusive rights — the cornerstone of the U.S. Copyright Act — are what incentivize massive investment in film and television. Midjourney, the plaintiffs allege, is exploiting those rights by training on “illegal copies” and encouraging subscribers to generate outputs featuring these characters “in every imaginable scene.”

This isn’t just about Warner Bros. Discovery’s catalog. The Motion Picture Association has endorsed the suit, warning that unchecked infringement threatens an industry that supports over two million jobs across all 50 states. And Warner Music Group has voiced similar concerns about AI systems that can generate music mimicking protected songs and artists.

The message is clear: copyright exposure isn’t just about what data went in. Liability now hinges on what comes out — and whether those outputs substitute for, or dilute the market value of, protected creative works.

What It Means: No More Hurry Up and Wait

These cases reveal two sides of the same coin:

For companies and legal teams: the compliance burden is immediate. Audit supply chains, scrutinize vendor contracts, and prepare for output claims. Don’t assume indemnities are meaningful or that “publicly available” data equals lawful use.
For creators and rights holders: the path forward lies in outputs and market harm. Document sales losses, track “sham” AI-generated works, and organize around collective licensing models.

The message is clear: waiting is no longer an option.

The dominoes are falling. What matters is how fast you act before the next one tips.

Lili Kazemi is General Counsel and AI Policy Leader at Anant Corporation, where she advises on the intersection of global law, tax, and emerging technology. She brings over 20 years of combined experience from leading roles in Big Law and Big Four firms, with a deep background in international tax, regulatory strategy, and cross-border legal frameworks. Lili is also the founder of DAOFitLife, a wellness and performance platform for high-achieving professionals navigating demanding careers.

Follow Lili on LinkedIn and X

🔍 Discover What We’re All About

At Anant, we help forward-thinking teams unlock the power of AI—safely, strategically, and at scale.
From legal to finance, our experts guide you in building workflows that act, automate, and aggregate—without losing the human edge.
Let’s turn emerging tech into your next competitive advantage.

Follow us on LinkedIn

👇 Subscribe to our weekly newsletter, the Human Edge of AI, to get AI from a legal, policy, and human lens.