Strike 3 Sues Meta for $359M: Adult Studios Allege Movie Gen AI Trained on Pirated Films

Meta is facing yet another AI training copyright lawsuit, but this one comes with a twist: the plaintiffs are adult film studios, the alleged training target is Meta's Movie Gen video model, and the evidence builds directly on Meta's own admissions from the earlier authors' litigation.

Strike 3 Holdings and Counterlife Media — the companies behind brands such as Vixen, Tushy, Blacked, and Deeper — filed the complaint in the U.S. District Court for the Northern District of California. They allege Meta downloaded at least 2,396 of their films through BitTorrent since 2018 to train its AI systems, and they are seeking statutory damages that could reach roughly $359 million under willful infringement multipliers.

This is the first major AI training lawsuit centered on video data sourced from pirate networks. It also tests a legal theory — contributory infringement through BitTorrent seeding — that a federal judge already allowed to proceed against Meta in the parallel authors' case earlier this year.

What the Complaint Actually Alleges

According to the complaint, Meta did not just download. It participated in BitTorrent swarms, which by design require uploading pieces of files to other peers in exchange for download speed. The plaintiffs frame that as a deliberate choice rather than an unavoidable side effect.

The core allegations:

• Meta downloaded at least 2,396 copyrighted adult films from pirate sources between 2018 and the present.
• The downloads were allegedly tied to training Meta Movie Gen (Meta's text-to-video model), Llama, and other AI systems that consume video.
• Meta allegedly used both corporate IP addresses owned by Facebook and "off-infra" stealth IPs linked to hidden datacenters to obscure the activity.
• Meta allegedly seeded the files back to other BitTorrent users, qualifying as direct distribution and contributory infringement.
• An unnamed Facebook employee allegedly downloaded content from a residential Comcast IP that correlates with the corporate and stealth IP activity.

The plaintiffs say their evidence comes from VXN Scan, an in-house BitTorrent monitoring tool Strike 3 has used in thousands of prior lawsuits against individual file sharers, cross-referenced against MaxMind's IP ownership database.

Why Movie Gen Changes the Calculus

Most prior AI copyright lawsuits have involved text (the New York Times, the authors class actions, Thomson Reuters v. Ross) or images (Getty, the Stability AI artists). Video training data has been a quieter front, partly because the models are newer and partly because the training corpora have been less publicly documented.

Movie Gen, announced by Meta in 2024 as a research model for high-fidelity video generation, makes the question concrete. The complaint argues that training on a curated catalog of professionally produced adult films could let Meta produce "Hollywood grade" generative video that competitors cannot replicate without similar source material.

That framing matters for the fourth fair use factor — market harm. In Bartz v. Anthropic and the early Meta books ruling, courts have leaned toward fair use when the AI output does not directly substitute for the training inputs. A video model that can generate output stylistically indistinguishable from the plaintiffs' catalog is a more direct market substitute than a chatbot that summarizes a novel.

The BitTorrent Seeding Theory Is Already Battle-Tested

The strongest part of this complaint may be that the legal theory is not novel. It piggybacks on developments in Kadrey et al. v. Meta, the authors class action filed by Richard Kadrey, Sarah Silverman, Christopher Golden, and others.

A short timeline of how seeding became a viable claim against Meta:

| Date | Development |

|---|---|

| Jan 2024 | Meta admits in discovery it used pirated book datasets (LibGen, Anna's Archive) for Llama training |

| June 2025 | Judge Vince Chhabria rules Meta's use of pirated books for training qualifies as fair use, but distribution claims survive |

| Dec 2025 | Authors seek to add a contributory infringement claim based on BitTorrent seeding |

| March 2026 | Judge Chhabria allows the seeding claim, calling counsel's delay excuse "lame" and "doubletalk" but granting the amendment |

| May 2026 | Strike 3 and Counterlife file the present suit, citing Meta's prior admissions |

In other words, Strike 3 is not asking the court to accept an untested theory. It is asking the court to apply the same theory that already survived dismissal in a different case against the same defendant.

How This Differs From Other Meta AI Cases

Three things set this case apart from the broader pile of Meta AI lawsuits:

1. The plaintiff is a litigation specialist. Strike 3 Holdings is the most prolific copyright plaintiff in the United States, filing thousands of suits a year against individual BitTorrent users. It has the infrastructure, the lawyers, and the appetite to litigate. It is not a publisher reluctantly stepping into court.

2. The content category is unique. Adult content is fully copyrightable in the United States, but the industry rarely participates in the high-profile AI policy debates that center on news publishers and book authors. A judgment here would extend the AI-training copyright case law into a new vertical.

3. The damages math is meaningful. At the maximum statutory rate of $150,000 per work for willful infringement, 2,396 films yield about $359 million. That is large enough to matter to Meta's quarterly numbers, and large enough to incentivize a settlement rather than a precedent-setting trial.

What Meta Is Likely to Argue

Meta has not yet responded to the complaint. Based on its filings in Kadrey, the playbook will probably include:

• Fair use on the training itself. The same argument that prevailed in the books case: training is transformative, the model does not output the works, and the market for the originals is not directly displaced.
• No proof of distribution of complete works. BitTorrent shares files in chunks. Plaintiffs typically struggle to prove that any single peer transmitted an entire work to any single recipient. (This is why the Kadrey plaintiffs pivoted to a contributory theory.)
• Challenges to VXN Scan. Defendants have repeatedly questioned the methodology of automated BitTorrent monitoring, and a well-funded defendant has more resources than the typical Doe defendant to litigate the tool's reliability.
• No connection between the IPs and AI training. Meta will probably argue that even if some Facebook IPs participated in BitTorrent swarms, plaintiffs cannot prove the downloads were for Movie Gen training rather than employee misconduct, security research, or unrelated uses.

What This Means for Other AI Companies

Even before any ruling, the filing changes the risk landscape:

• Video-generation models are now a litigation target. Sora 2, Runway, Pika, and any model trained on web-scraped video are exposed to similar theories.
• Acquiring training data via BitTorrent is the weakest legal posture. Even where training itself is found to be fair use, the act of using a peer-to-peer protocol that requires reciprocal seeding creates a separate, harder-to-defend distribution claim.
• Internal IP discipline matters. "Off-infra" stealth datacenters used to obscure activity become evidence of consciousness of guilt rather than effective concealment, especially once corporate IP correlations are documented.
• Specialty content categories may follow. If adult studios can sue, expect filings from independent film distributors, stock footage providers, niche documentary producers, and other specialty rightsholders whose catalogs were plausibly scraped.

For businesses building AI products, this suggests two practical adjustments. First, demand documentation from training data vendors that proves licensed or public-domain origin. Second, treat BitTorrent acquisition of any copyrighted material — even for legitimate research purposes — as a non-starter regardless of the underlying fair use defense for training.

What to Watch Next

• Meta's motion to dismiss. Expected within roughly 60 days of service. Whether Meta tries to consolidate this case with Kadrey, or argues it should be handled separately, will signal its strategy.
• VXN Scan challenges. Strike 3's monitoring tool has faced procedural challenges in Doe defendant cases but has rarely been deeply litigated by a defendant with Meta's resources.
• Discovery into Movie Gen's training corpus. If the case survives early dismissal, document discovery into Movie Gen's training data could reshape what we know about commercial video model training.
• Settlement signals. Strike 3 has a long history of out-of-court settlements. A quiet resolution would be consistent with its pattern but would deprive the field of a published opinion on AI video training.

Key Takeaways

• Strike 3 Holdings and Counterlife Media are suing Meta in the Northern District of California, alleging 2,396 adult films were downloaded via BitTorrent for AI training.
• Potential statutory damages reach approximately $359 million.
• The complaint targets Meta's Movie Gen video model, Llama, and other AI systems consuming video data.
• The seeding-as-contributory-infringement theory has already survived dismissal in the parallel Kadrey authors case.
• This is the first major AI copyright lawsuit focused specifically on video training data and on adult content, opening a new front in AI copyright litigation.
• For AI developers, the case reinforces that fair use defenses for training do not insulate the acquisition method from separate infringement liability.

This article is for informational purposes only and is not legal advice. Allegations in the complaint have not been proven in court. For specific legal questions about AI copyright exposure, consult a qualified attorney. Sources: Strike 3 Holdings & Counterlife Media v. Meta Platforms (N.D. Cal., May 2026); Kadrey et al. v. Meta Platforms; U.S. Copyright Office.