Scott Turow Sues Meta Over Llama AI: Legal Breakdown

HOST

From DailyListen, I'm Alex. Scott Turow, the guy behind legal bestsellers like Presumed Innocent, just jumped into a real courtroom drama. He's teaming up with five big publishers—Elsevier, Cengage, Hachette, Macmillan, McGraw Hill—to sue Meta and Mark Zuckerberg personally. They say Meta pirated millions of copyrighted books and articles from sites like LibGen and Anna's Archive to train its Llama AI models, all without permission or payment. Filed Tuesday in Manhattan federal court, case number 26-cv-3689. This could shake up how AI companies build their tech. We're joined by Catherine, our legal analyst, to break down the claims and what's at stake.

CATHERINE

The court filing in the Southern District of New York accuses Meta of straight copyright infringement under 17 U.S.C. sections 106 and 501. Plaintiffs say defendants torrented millions of works, including Turow's novels and publishers' journals, to build Llama. They even claim Zuckerberg personally greenlit it, pointing to his "move fast and break things" motto. The suit adds removal of copyright management info as a claim. Cause leads to effect here: no license, no payment, direct copies from pirate sites. That's the core allegation. Courts look first at reproduction rights, and this complaint hammers unauthorized copying as the violation.

HOST

Hold on—Zuckerberg personally authorized grabbing books from pirate libraries? That feels like naming the CEO to make a point. How rare is pinning the boss like that in these suits?

CATHERINE

Plaintiffs name Zuckerberg to pierce the corporate veil, alleging he directed the infringement. The complaint quotes internal pushes for speed over legality. This sets it apart from standard suits against companies alone. Precedent from cases like New Era v. Rice shows executives can face liability if they actively encourage violations. Here, they tie it to his emails and strategy calls. But success depends on discovery proving his direct hand. Meta's response so far? Silence in public—no filing yet since it's brand new, filed Tuesday. They'll likely move to dismiss, arguing fair use.

HOST

Fair use—Meta's go-to shield. Judge Chhabria ruled last June that training AI on books can be fair use if no market harm shows. Does this suit dodge that by showing harm?

CATHERINE

Judge Chhabria's June 2025 ruling in a similar case granted Meta summary judgment. He said plaintiffs lacked evidence their books' use harmed the market. But this complaint flips it. Plaintiffs highlight a licensing market Meta knows about—they signed deals in 2022 with four African-language publishers for training data, plus pacts with Fox, CNN, USA Today. That's binding evidence under fair use's fourth factor: effect on potential market. Courts weigh if licenses exist and defendant bypassed them. Chhabria noted weak harm proof before; here, named deals show priced options Meta ignored for U.S. books. Stronger footing.

HOST

Those Meta licenses with news outlets and others—it's like they pay for some data but snatch books for free. Puts real pressure on the fair use claim.

CATHERINE

Exactly. Fair use doctrine from 17 U.S.C. 107 balances four factors, starting with purpose—transformative use favors tech firms. But market harm tips against if licenses prove value. Meta's statement pushes back: "Courts have rightly found training AI on copyrighted material can qualify as fair use," citing productivity gains. They argue Llama creates new tools, not copies books verbatim. No output means no direct competition, they say. Yet plaintiffs counter with Amazon's flood of AI books—one writer dropped three in three months, even leaving in prompts to mimic real authors. That shows real dilution.

HOST

Turow calls it "stolen words." With pirate sites named, does the suit prove Meta knew it was illegal?

CATHERINE

The complaint details Meta scraping LibGen and Anna's Archive—sites sued repeatedly for piracy. Plaintiffs say Meta knew, chose speed. Effect: millions reproduced, distributed internally for Llama training. Precedent like MAI v. Peak shows even temporary copies count as infringement. Zuckerberg's role amps it—they allege he encouraged despite legal flags. But Meta defends by noting AI needs vast data; public domain plus fair use covers it. No procedural updates yet—no response filed, no hearings set. Early days, docket just opened.

HOST

Broader fight—six authors sued Anthropic, OpenAI, Google, Meta, xAI, Perplexity just days ago on October 4th. How does Turow's case fit the pile-up?

CATHERINE

This opens a new front against Llama specifically, but mirrors suits like the New York Times against OpenAI. Plaintiffs there showed verbatim outputs; here, focus is training ingestion. Anthropic faces individual author claims for Claude training. Common thread: all allege unauthorized copies for LLMs. Chhabria's ruling slowed some, but these build market-harm evidence post-2025. Tech side holds fair use wins in cases like Google Books—scanning for search was okay. Difference? Publishers now tout licenses as proof of harm, forcing courts to weigh opted-in markets Meta skipped.

HOST

Meta's quiet on response, but they've done these deals before. Any hint they'll settle like with news orgs?

CATHERINE

No settlement signals yet—too fresh. Meta settled licensing with Associated Press, others, but fought publishers hard pre-Chhabria. Procedural next: Meta files answer or motion to dismiss in weeks, likely arguing fair use and no Zuckerberg liability. Plaintiffs seek class status for all affected authors via S.C.R.I.B.E., Inc. Precedent favors certification if common issues dominate. But defenses point to Llama's open weights—unlike closed models, users adapt it freely. That undercuts harm claims, they argue, since outputs aren't Meta-controlled.

HOST

Open weights on Llama mean anyone tweaks it. Does that weaken the publishers' market-harm angle?

CATHERINE

It bolsters Meta's fair use pitch. Courts ask if defendant's use supplants the original market. Llama's models are downloadable, fine-tunable—thousands do it daily per Meta's LlamaCon 2025 stats. Plaintiffs must prove training itself tanks book sales or licensing. They cite AI books on Amazon mimicking styles, but correlation isn't causation. Counterpoint: Turow's own sales hold steady, per public charts. Still, class action scale—millions of works—could force talks if discovery uncovers pirate proofs.

HOST

Turow's no stranger to courtrooms—he writes them. This feels personal with his books in the mix. What's the real risk to AI builders if they lose?

CATHERINE

Loss means injunctions halting Llama use without licensed data, plus damages per work—potentially billions. Precedent from Napster shut down peer-to-peer for similar copying. Effect cascades: other firms like Anthropic pause training, negotiate blanket licenses. Publishers gain leverage; Authors Guild pushed similar post-2023. But win for Meta reinforces fair use for AI, greenlighting data hunger. Ongoing suits against OpenAI show split courts—some deny dismissal, others grant. This one's key for executive liability too; Zuckerberg walking free sets bar low.

HOST

Billions in damages from millions of books—that's the nightmare scenario. But Meta's motto was "move fast and break things." Did they break copyright law knowingly?

CATHERINE

Complaint says yes—internal docs allegedly show awareness of pirate risks, Zuckerberg overrides. They tie to February 2026 trial where he testified on platform harms, but that's separate. Meta's view: innovation demands data access; fair use evolved for tech like search engines. No final ruling here, but Chhabria's words linger: insufficient harm evidence dooms claims. Plaintiffs bet their licenses prove it. Watch for Meta's first filing—it'll frame the battle.

HOST

Wild to think one author's thriller turns real. Catherine, spot on as always—facts cut through the noise. Folks, that's the latest on Turow versus Meta: stolen books fueling Llama, fair use on trial, Zuckerberg in the hot seat. Check court dockets for updates. I'm Alex. Thanks for listening to DailyListen.