Originally published at twarx.com - read the full interactive version there.
Last Updated: June 21, 2026
A federal judge has now ruled that a porn company can sue Meta for torrenting its adult films for AI training, and that single decision just torched the 'fair use' argument every AI company has leaned on — not via a sympathetic plaintiff, but via a porn studio with a track record of suing individual downloaders for thousands of dollars. If Meta cannot escape a lawsuit for torrenting adult films to feed its AI, no company's training dataset is legally safe.
On June 11, 2026, U.S. District Judge Eumi K. Lee denied Meta's motion to dismiss a copyright lawsuit brought by Strike 3 Holdings and Counterlife Media — the owners of premium adult brands Blacked, Vixen, and Tushy — over the alleged torrenting of more than 2,300 films to train Meta's AI models. The companies are seeking up to $359 million in damages.
After reading this, you will understand exactly what survived dismissal, why BitTorrent makes Meta uniquely exposed, and what this means for every AI lab quietly trained on scraped data. For broader context on how the law is catching up to model builders, see our running coverage of AI copyright law.
The Strike 3 Holdings v. Meta ruling marks the first time a court has let an AI training copyright case proceed specifically over BitTorrent-based acquisition. Source: Mashable / Marcin Golba NurPhoto via Getty Images
Coined Framework
The Torrent Liability Threshold — the legal inflection point at which AI data acquisition crosses from passive scraping into active copyright distribution, fundamentally shattering the fair use defence that every major AI company currently relies upon
It names the precise moment an AI lab stops being a mere consumer of data and becomes a distributor of it. Once that line is crossed, the four-factor fair use analysis collapses — because you cannot 'transformatively use' something you are simultaneously re-uploading to strangers.
What Was Ruled: The Exact Court Decision and Key Facts
This is the single most consequential fact: a federal judge has now formally accepted that an AI company can be sued not just for using copyrighted material, but for distributing it via the very mechanism used to acquire it.
Which judge issued the ruling and when
On June 11, 2026, U.S. District Judge Eumi K. Lee filed the order denying Meta's motion to dismiss, according to Mashable's reporting by Anna Iovine. Lee found that Strike 3 Holdings and Counterlife Media 'have plausibly alleged that [Meta] is liable for direct, vicarious, and contributory copyright infringement based on the torrenting of their films.'
What Meta's motion to dismiss argued — and why it failed
Meta filed its motion to dismiss in October 2025, calling the claims 'nonsensical and unsupported' and arguing the porn downloads were for 'personal use' by individuals. Judge Lee was unconvinced. She pointed to IP addresses tracing back to Meta's corporate offices torrenting similar files with identical names, all in a single day, ranging from cartoons to porn.
'It strains credulity to suggest that these correlations are mere coincidence and the product of individual human selections.' — U.S. District Judge Eumi K. Lee
The specific legal claims that survived dismissal
Three theories of liability survived: direct infringement (Meta downloaded the works), vicarious infringement (Meta benefited and had the right to control the conduct), and contributory infringement (Meta materially contributed to infringement by others). Critically, the suit alleges the corporate IP addresses acted 'consistently in non-human patterns' involving 'mass infringement beyond what a human could consume.' For a primer on how U.S. courts weigh these doctrines, the exclusive rights in Section 106 are the statutory foundation.
To be precise about the stakes: this is a denial of a motion to dismiss, not a final verdict. It means the plaintiffs cleared the lowest bar — pleading a plausible claim — and the case now proceeds to discovery. But that procedural milestone forces Meta to produce internal engineering records it has fought hard to keep sealed.
2,300+
Copyrighted adult films allegedly torrented (2018–2025)
[Mashable, 2026](https://mashable.com/tech/porn-company-can-sue-meta-torrenting-copyright)
$359M
Maximum damages sought by the plaintiffs
[Mashable, 2026](https://mashable.com/tech/porn-company-can-sue-meta-torrenting-copyright)
Jun 11, 2026
Date Judge Lee denied Meta's motion to dismiss
[Mashable, 2026](https://mashable.com/tech/porn-company-can-sue-meta-torrenting-copyright)
Who Is Strike 3 Holdings and Why This Plaintiff Is Uniquely Dangerous for Meta
Most AI copyright plaintiffs are authors, news publishers, or artists — sympathetic, but litigation novices. Strike 3 is the opposite: a battle-hardened serial litigant with forensic infrastructure built specifically to track BitTorrent activity.
Strike 3's business model: from suing individuals to targeting Big Tech
Strike 3 Holdings owns several popular porn sites including Blacked, and per 404 Media's reporting, its portfolio spans Vixen, Tushy, and Deeper. For years its core business has been filing thousands of lawsuits against individual BitTorrent users — making it one of the most prolific copyright litigants in U.S. federal court history.
Strike 3 has filed thousands of individual BitTorrent suits — meaning it arrives in the Meta case with pre-built forensic IP-tracking pipelines that most plaintiffs would need to spend years and millions building from scratch.
The 'copyright troll' history that paradoxically strengthens its standing
Critics have long labelled Strike 3 a 'copyright troll' for pursuing small-scale downloaders. But that same history is now a weapon: the company has refined its evidentiary methods through years of litigation, knows exactly how to authenticate BitTorrent swarm data, and has zero fear of Meta's legal budget. The irony — a company built on suing individuals now weaponising the identical legal theory against a trillion-dollar firm — is precisely what gives it credibility on the technical mechanics.
Counterlife Media's role as co-plaintiff
Counterlife Media, in which Strike 3 holds a majority ownership interest, is the lesser-known co-plaintiff. It adds additional copyrighted works to the damages pool, expanding the combined claim toward the $359 million figure cited in the complaint.
Unlike web scraping, BitTorrent makes every downloader a simultaneous uploader — the mechanism at the heart of the Torrent Liability Threshold.
How Meta Allegedly Torrented and Seeded Adult Films: The Technical Mechanics
To understand why this case is different from every other AI copyright suit, you have to understand how BitTorrent actually moves data.
How BitTorrent works and why seeding is legally distinct from downloading
BitTorrent is a peer-to-peer protocol. Instead of pulling a file from one central server, your client downloads small fragments from dozens of peers — and simultaneously uploads the fragments you already have to other peers. That uploading is called 'seeding.' By design, you cannot meaningfully participate in a swarm as a pure consumer; the protocol's efficiency depends on you redistributing. The official BitTorrent protocol specification confirms this reciprocal upload mechanic is foundational, not optional.
The Torrent Liability Threshold: How Acquisition Becomes Distribution
1
**Corporate IP joins the swarm**
An IP address tracing to Meta's offices connects to a BitTorrent swarm hosting a copyrighted film. Latency: seconds. This is the acquisition intent.
↓
2
**Fragments download (infringement #1)**
The client pulls file pieces from peers. This is classic direct infringement — reproduction of a copyrighted work.
↓
3
**Fragments upload to peers (infringement #2)**
Simultaneously, the client seeds owned pieces back to the swarm — distributing the work to unknown third parties. This crosses the Torrent Liability Threshold.
↓
4
**Forensic capture by Strike 3**
Strike 3's monitoring infrastructure records the corporate IP both downloading and seeding — generating timestamped distribution evidence.
↓
5
**Films ingested into the AI training pipeline**
The downloaded media is fed into model training corpora — the original purpose. But the distribution at step 3 already created independent liability.
The sequence matters because step 3 — seeding — creates liability that no amount of 'transformative use' argument at step 5 can undo.
What 'active seeding' means for Meta's liability exposure
This is the legal nuclear option. Under U.S. copyright law, the exclusive right to distribute is separate from the right to reproduce. A defendant who only downloads is an infringer; a defendant who also uploads is a distributor — and distribution typically supports the contributory and vicarious theories that survived in this case. It is no accident Judge Lee accepted all three liability theories.
The scale: nearly 2,400 films across multiple premium platforms
The complaint alleges roughly 2,300-plus films were torrented between 2018 and 2025. The IP addresses, per the suit, 'acted consistently in non-human patterns' — mass infringement 'beyond what a human could consume.' That phrasing is deliberate: it converts the scale itself into evidence of a systematic, programmatic operation rather than employee misconduct.
The fatal flaw in Meta's defence isn't that it downloaded porn. It's that BitTorrent forced Meta to redistribute it — and you cannot claim 'transformative fair use' on a file you're simultaneously uploading to strangers.
The AI Training Data Copyright Crisis: How This Case Fits the Broader Landscape
This ruling does not exist in a vacuum. It lands in the middle of a multi-front legal war over how foundation models are fed. Our deep dive on AI training data sourcing tracks how this war is reshaping the entire model-building supply chain.
Timeline of major AI copyright lawsuits
Meta already faces parallel litigation from authors in the Kadrey v. Meta case, alongside suits against OpenAI from the New York Times and authors including John Grisham, plus music-publisher claims against Anthropic over song lyrics in Claude training data. The Strike 3 case is the newest front — and the one with the most dangerous mechanism.
Why AI companies believed fair use protected them
The prevailing theory has been transformative fair use under Section 107 of the Copyright Act. The argument: training a model extracts statistical patterns, not the expressive content itself, so the use is transformative. Per Mashable, Strike 3 and Counterlife only became aware of Meta's BitTorrent activity through press coverage of the January 2025 books lawsuit — discovery in which revealed Meta had pirated books for AI training.
Notably, Meta won that books case in June 2025 — but the judge wrote that the plaintiffs might have succeeded with different legal arguments, explicitly leaving the door open for suits exactly like this one.
Coined Framework
The Torrent Liability Threshold in practice
When a model builder scrapes a public webpage, it ingests but does not redistribute. When it joins a torrent swarm, the protocol forces redistribution — pushing the conduct across the threshold from defensible ingestion into indefensible distribution.
The judge who ruled for Meta in the June 2025 books case effectively wrote the roadmap for the porn case that now threatens it — by signalling that distribution-based theories could succeed where pure-use theories failed.
Discovery in Strike 3 v. Meta could expose the full architecture of how Llama-class models source their training data — the most consequential document production in AI legal history.
What Is It: A Plain-Language Explanation for Non-Experts
Strip away the legalese. A company that makes adult films found evidence that Meta's office computers used a file-sharing tool — BitTorrent — to download more than 2,300 of its movies between 2018 and 2025, allegedly to help train Meta's AI. The film company sued. Meta tried to get the case thrown out, claiming it was just employees downloading porn for personal use. The judge said that excuse doesn't hold up given the sheer volume and machine-like pattern, and let the lawsuit continue.
Why should a small-business owner care? Because the case tests a question that affects every AI tool you use: can companies legally train AI on content they didn't pay for? If the answer becomes 'no,' the cost of building AI rises for everyone — and the price you pay for AI products could follow.
How It Works: The Mechanism in Plain Language
Think of BitTorrent like a potluck dinner where everyone is required to both take food and bring food. You can't just show up and eat — the moment you grab a plate, you're also serving others. That's the difference between BitTorrent and a normal website download.
Before vs After: Web Scraping vs Torrenting for AI Training
A
**Web Scraping (the old model)**
A crawler like Common Crawl reads a public page and copies the text. One-way. The AI company is a consumer only. Fair use is at least arguable.
↓
B
**Torrenting (the Meta allegation)**
The AI company joins a swarm, downloads fragments AND uploads them to strangers. Two-way. Now it is a distributor — the legally fatal distinction.
↓
C
**The legal consequence**
Web scraping faces one infringement theory (reproduction). Torrenting faces three (direct, vicarious, contributory) — exactly what survived in this case.
The architecture of acquisition — not the use of the data — is what determines legal exposure here.
For builders running their own data pipelines, the lesson maps directly onto modern AI stacks. If you orchestrate ingestion through automated workflows or a RAG pipeline, the method of acquisition now carries independent legal weight — separate from how you use the data downstream.
Meta's Legal Defence Strategy and Why It Failed at Dismissal Stage
The 'rogue employee' argument and why the court rejected it
Meta's central defence was that any unauthorised torrenting was the work of rogue individuals acting outside company policy. Judge Lee found this implausible given the volume — files with identical names, downloaded the same day, spanning cartoons to porn from corporate IP addresses. As she wrote, the correlations were unlikely to be 'mere coincidence and the product of individual human selections.'
Fair use as an AI training defence: its limits here
The four-factor fair use test under Section 107 weighs purpose, nature of the work, amount used, and market effect. Adult studios can argue all four cut against Meta — particularly market harm, since AI-generated adult content directly competes with human-produced films. And crucially, the seeding conduct undermines the 'purpose' factor: redistribution is not a transformative research purpose.
What surviving a motion to dismiss actually means
A denied motion to dismiss sends the case to discovery — the phase where Meta must produce internal communications, engineering documentation, Slack messages, and data-pipeline records. This is where the real danger lives: discovery could expose precisely how Meta sourced training data at scale.
❌
Mistake: Treating 'fair use' as a blanket shield for any acquisition method
Many teams assume that because model training is transformative, how they obtained the data is irrelevant. The Torrent Liability Threshold proves otherwise — the acquisition method created independent distribution liability.
✅
Fix: Audit your ingestion method, not just your use case. Prefer licensed datasets or one-way crawls (Common Crawl) over any peer-to-peer protocol.
❌
Mistake: Assuming corporate IP activity is invisible
Meta's downfall was IP addresses tracing to its corporate offices being captured in BitTorrent swarms. Forensic monitoring firms record this in real time.
✅
Fix: Maintain a documented data-provenance log for every training source. If you can't show where a dataset came from, you can't defend it.
❌
Mistake: Relying on the 'rogue employee' deflection
At scale, the volume itself becomes evidence of systematic conduct. Judge Lee rejected this exact argument as straining credulity.
✅
Fix: Implement enforceable data-acquisition policies with technical controls (blocked torrent clients on corporate networks) rather than relying on after-the-fact denials.
How to Use This Knowledge: A Worked Compliance Demonstration
Here's a concrete, runnable example of how a builder should screen a training corpus for the Torrent Liability Threshold before ingesting it. The principle: classify every source by acquisition method and licence status.
python — data-provenance gate
Provenance gate: block any source acquired via P2P/torrent
Run this BEFORE adding a dataset to your training pipeline.
ALLOWED_METHODS = {'licensed', 'public_crawl', 'first_party', 'opt_in'}
BLOCKED_METHODS = {'bittorrent', 'p2p_seed', 'unknown'}
def screen_source(source):
# source = {'name':..., 'method':..., 'license':...}
method = source['method']
if method in BLOCKED_METHODS:
return f"BLOCKED: {source['name']} crossed the Torrent Liability Threshold"
if method in ALLOWED_METHODS and source['license']:
return f"CLEARED: {source['name']} ({source['license']})"
return f"REVIEW: {source['name']} — provenance incomplete"
corpus = [
{'name':'Common Crawl shard 42','method':'public_crawl','license':'CC-derived'},
{'name':'Vendor licensed images','method':'licensed','license':'commercial'},
{'name':'Leaked film torrent','method':'bittorrent','license':None},
]
for s in corpus:
print(screen_source(s))
Actual output:
output
CLEARED: Common Crawl shard 42 (CC-derived)
CLEARED: Vendor licensed images (commercial)
BLOCKED: Leaked film torrent crossed the Torrent Liability Threshold
This is the kind of guardrail teams now bake into ingestion. To automate provenance gating across a full pipeline, you can wire checks like this into an orchestration layer or explore our AI agent library for ready-made compliance agents. Builders standardising on multi-agent systems can assign a dedicated provenance-auditing agent — and you can deploy a prebuilt data-compliance agent from Twarx using frameworks like LangGraph or n8n.
Comparison: Meta's AI Copyright Cases vs Competitors
The defining variable across these cases is acquisition method. Web crawling is defensible; peer-to-peer seeding is not.
CompanyModelPrimary acquisition methodDistribution liability?Key plaintiff
MetaLlamaBitTorrent (alleged)Yes — direct, vicarious, contributoryStrike 3 Holdings
OpenAIGPTWeb crawl / Common CrawlNo — reproduction onlyNew York Times
GoogleGeminiWeb index crawlNo — strongest fair use postureMultiple jurisdictions
AnthropicClaudeWeb scrapingNo — scraping, not P2PMusic publishers
Stability AIStable DiffusionScraped image datasetsNo — reproduction theoryGetty Images
The takeaway is stark: Meta is the only major lab accused of using a protocol that, by design, makes the downloader a simultaneous distributor. OpenAI's NYT case, Anthropic's lyrics suit, and Stability's Getty Images litigation all turn on reproduction. Meta's case turns on reproduction and distribution.
What It Means for Small Businesses
For a small business, this ruling cuts two ways.
The risk: If you fine-tune open models or build products on scraped datasets, the acquisition method of your training data is now a liability surface. A cheap dataset of dubious origin could become an expensive lawsuit.
The opportunity: If you own original content — photography, video, written work, proprietary data — its licensing value is rising. As fair use defences narrow, AI companies are signing more licensing deals, and your archive becomes a sellable asset. For practical guidance, see our playbook on content licensing for the AI era.
If statutory damages hit the wilful-infringement maximum of $150,000 per work across roughly 2,400 films, the arithmetic lands almost exactly at the $360M the suit claims — which is why the damages figure isn't rhetorical, it's calculated.
Who Are Its Prime Users (Who This Ruling Most Affects)
AI/ML engineers and data leads building training corpora at mid-to-large companies — they now own provenance liability.
Content owners (studios, publishers, photographers, stock libraries) — newly empowered to demand licensing.
Legal and compliance teams at AI startups (10–500 employees) — must add data-acquisition audits.
AI policy researchers and journalists tracking the fair use erosion.
Licensing marketplaces like Shutterstock, Getty, and Spawning.ai — direct commercial beneficiaries.
When to Use It (and When Not To)
Apply the Torrent Liability Threshold framework when: you are sourcing any third-party media for model training, fine-tuning, or RAG; you are acquiring datasets of unknown provenance; or you are evaluating an acquisition target's AI assets in due diligence.
It's less relevant when: you train exclusively on first-party data, fully licensed corpora, or opt-in user data. In those cases the acquisition method is clean and the distribution-liability concern evaporates. Alternatives like licensed datasets cost more upfront but eliminate the tail risk that this ruling just made concrete. Teams comparing approaches should review our breakdown of fine-tuning vs RAG architectures.
Industry Impact: What This Ruling Means for AI Training Data Practices
Immediate implications for dataset builders
If Strike 3 reaches trial and wins, statutory damages of $150,000 per wilful infringement across roughly 2,400 works yield a theoretical maximum near $360 million — matching the complaint. That arithmetic is now a template other rightsholders will copy.
Could this trigger audits of existing model histories?
Almost certainly. Labs that trained on opaque corpora face a new question in every due-diligence and investor conversation: where did the data come from? Expect retroactive provenance audits to become standard. Teams modernising their stack should review our guide to building AI compliance into production systems.
The content licensing market: who benefits
OpenAI, Google DeepMind, and Mistral have already begun signing licensing deals with publishers and media companies. This ruling accelerates that shift. Platforms like Getty Images (itself suing Stability AI), Shutterstock, and Spawning.ai stand to gain if courts require licensed training data.
$150K
Max statutory damages per wilful infringement (US copyright law)
[U.S. Copyright Act §504](https://www.copyright.gov/title17/92chap5.html#504)
3
Liability theories that survived dismissal (direct, vicarious, contributory)
[Mashable, 2026](https://mashable.com/tech/porn-company-can-sue-meta-torrenting-copyright)
2018–2025
Period of alleged torrenting activity
[Mashable, 2026](https://mashable.com/tech/porn-company-can-sue-meta-torrenting-copyright)
[
▶
Watch on YouTube
How AI training data copyright lawsuits could reshape the industry
AI policy & copyright explainer
](https://www.youtube.com/results?search_query=Meta+AI+copyright+lawsuit+torrenting+explained)
Expert and Community Reactions to the Meta Porn Torrenting Ruling
Legal experts on surviving the motion to dismiss
404 Media, which surfaced detailed coverage of the underlying allegations, has emphasised that the court's 'strains credulity' language is unusually pointed judicial scepticism — a hostile signal for Meta's future arguments. Commentary at The Verge has echoed that the distribution angle is the genuinely novel threat.
AI ethics community: data provenance and consent
AI ethics researchers point to this case as evidence the field needs mandatory training-data provenance standards — the 'nutrition label for datasets' analogy that recurs in AI governance discussions. The Electronic Frontier Foundation has long warned that acquisition method matters as much as use.
Tech community reaction
Across legal communities, commentators note that the seeding behaviour is the aspect most likely to defeat fair use — because it demonstrates distribution, not merely research. And the broader irony has not gone unnoticed: a company that built its business suing individuals for downloading films is now wielding the same mechanism against a company with a market cap in the trillions.
The adult industry has always been first to the digital-copyright frontier — from VHS to streaming DRM. It is fitting, and a little terrifying for Big Tech, that it's now first to the AI training-data frontier too.
Good Practices: Best Practices and Common Pitfalls
Do maintain a written provenance record for every training source — name, method, licence, date.
Do block torrent and P2P clients on corporate networks via technical policy, not just HR rules.
Do prefer licensed datasets or one-way crawls; budget for licensing as a cost of doing AI business.
Don't rely on 'transformative use' to excuse a tainted acquisition method.
Don't assume corporate IP activity is private — forensic monitoring captures swarm participation in real time.
Don't deploy the 'rogue employee' defence; at scale, volume itself becomes evidence of systematic conduct.
Average Expense to Use It: The Real Cost of Compliance
What does building a defensible training pipeline actually cost? A realistic breakdown:
Free tier: Public datasets like Common Crawl and openly licensed corpora cost $0 to acquire, but require legal review.
Licensing: Stock libraries and publisher deals range widely — from per-asset fees to seven- and eight-figure enterprise licences for AI training rights.
Provenance tooling: Building an automated gate (like the demonstration above) is a few engineer-days; ongoing audit tooling can run as a modest monthly SaaS line item.
The hidden cost — litigation: The downside of skipping all this is the Strike 3 template: up to $150,000 per work in statutory damages. For 2,400 works, that's the $359M figure in this very case.
The math is brutal but clear: compliance is cheap relative to the tail risk. For teams automating ingestion through enterprise AI workflows, provenance gating is the single highest-ROI control you can add.
What Comes Next: Case Timeline, Discovery, and Broader Consequences
Expected next steps in Strike 3 Holdings v. Meta
With dismissal denied, the case enters discovery. Meta must produce internal documents, Slack messages, engineering specs, and data-pipeline logs revealing the scope of its BitTorrent-based acquisition. This is where the case becomes existentially dangerous — not because of this lawsuit alone, but because the discovered documents could fuel every other copyright suit Meta faces.
How discovery could reshape the debate
If internal records show systematic, sanctioned torrenting, the 'rogue employee' narrative dies permanently — and the precedent extends to authors, publishers, and musicians.
Legislative and regulatory responses
The EU AI Act, with its staged training-data transparency requirements, creates parallel pressure on Meta's European operations. Congressional proposals touching the DMCA and the No Fakes Act are being watched closely by litigants on both sides.
2026 H2
**Discovery battles intensify**
Expect Meta to fight document production aggressively. The June 2026 ruling's pointed language suggests the bench is unsympathetic — raising the odds discovery proceeds broadly.
2027
**Settlement pressure mounts**
Given the $359M exposure and reputational risk, a settlement becomes plausible — mirroring how AI labs have increasingly chosen licensing deals over courtroom risk.
2027–2028
**Provenance becomes table stakes**
As distribution-based theories prove durable, mandatory training-data provenance — already pushed by the EU AI Act — becomes standard industry practice.
The likely trajectory: from contested fair use defences toward mandatory licensing and provenance standards across the generative AI industry.
Frequently Asked Questions
Why can a porn company sue Meta for torrenting its adult films for AI training, and what did the judge rule?
On June 11, 2026, U.S. District Judge Eumi K. Lee denied Meta's motion to dismiss, ruling that a porn company can sue Meta for torrenting its adult films for AI training — allowing Strike 3 Holdings and Counterlife Media to proceed. Lee found the plaintiffs 'plausibly alleged' Meta is liable for direct, vicarious, and contributory copyright infringement based on torrenting their films. She rejected Meta's argument that the activity was personal-use downloading by rogue employees, writing it 'strains credulity' given that corporate IP addresses torrented identically named files the same day, from cartoons to porn. Importantly, this is a denial of dismissal — not a final verdict. It means the case now advances to discovery, where Meta must produce internal records. The plaintiffs seek up to $359 million in damages over more than 2,300 allegedly torrented adult films.
How did Meta allegedly use BitTorrent to download adult films for AI training?
According to the complaint, IP addresses tracing back to Meta's corporate offices used the BitTorrent program to download more than 2,300 copyrighted adult films between 2018 and 2025. The suit alleges these addresses 'acted consistently in non-human patterns' involving 'mass infringement beyond what a human could consume' — suggesting a systematic, programmatic operation rather than individual viewing. BitTorrent is a peer-to-peer protocol: users simultaneously download file fragments and upload (seed) them to other peers. That dual behaviour is central to the case, because seeding means Meta wasn't merely consuming the films — it was redistributing them to third parties, which supports the distribution-based infringement theories that survived dismissal.
What is Strike 3 Holdings and who owns Vixen, Tushy, and Blacked?
Strike 3 Holdings is a porn holding company that owns several popular adult brands, including Blacked, Vixen, Tushy, and Deeper, per reporting by Mashable and 404 Media. It is one of the most prolific copyright litigants in U.S. federal court history, having filed thousands of lawsuits against individual BitTorrent users — a practice that earned it the 'copyright troll' label. That extensive litigation history is exactly why it's dangerous to Meta: Strike 3 arrives with battle-tested legal strategy and forensic IP-tracking infrastructure. Counterlife Media, in which Strike 3 holds a majority ownership interest, is the co-plaintiff, adding more copyrighted works to the combined damages claim of up to $359 million.
How much money is Meta being sued for in the adult film copyright case?
Strike 3 Holdings and Counterlife Media are seeking damages of up to $359 million. That figure is not arbitrary — it reflects U.S. copyright law's statutory damages, which can reach $150,000 per wilful infringement. Multiply that maximum across roughly 2,400 allegedly infringed films and the arithmetic lands close to the $360 million claimed. Whether the plaintiffs ultimately recover anything near that depends on proving wilful infringement and surviving the full trial — this ruling only cleared the motion-to-dismiss stage. Still, the damages template is significant: it's the same calculation other rightsholders (authors, publishers, musicians) could apply to their own AI training-data claims against Meta and other labs.
Does using copyrighted content for AI training count as fair use?
It depends — and that's the unsettled core of AI copyright law. The dominant defence is transformative fair use under Section 107 of the Copyright Act, arguing that training extracts statistical patterns rather than reproducing expressive content. In June 2025, Meta won a books-based case partly on this theory — but the judge noted plaintiffs might have won with different arguments. The Strike 3 case is harder for Meta because the acquisition method (BitTorrent seeding) involves distribution, which fair use does not cleanly protect. Adult studios can also argue market harm, since AI-generated adult content competes with human-produced films. The bottom line: fair use may shield certain transformative uses, but it does not automatically excuse how the data was obtained — especially when obtaining it involves redistributing it.
Why is torrenting legally worse than web scraping for AI training data purposes?
Web scraping is a one-way operation: a crawler reads and copies content. The AI company is purely a consumer, so liability is limited to the reproduction right — and fair use is at least arguable. BitTorrent is two-way by design: while you download fragments, you also upload (seed) them to other peers. That makes you a distributor, not just a consumer. Under U.S. copyright law, the distribution right is separate from the reproduction right, and distribution supports vicarious and contributory infringement theories — all three of which survived in the Meta case. This is the Torrent Liability Threshold: the acquisition method itself crosses from defensible ingestion into active copyright distribution, undercutting any fair use claim. That's why no major lab using only web crawling faces the same exposure.
What does this ruling mean for other AI companies like OpenAI and Google?
The ruling primarily threatens companies whose data acquisition involved peer-to-peer distribution — which uniquely describes Meta's alleged BitTorrent use. OpenAI (sued by the New York Times), Google (Gemini), Anthropic (Claude), and Stability AI mostly relied on web crawling or scraped datasets, exposing them to reproduction claims but not distribution liability. However, the broader signal matters: courts are increasingly sceptical of blanket fair use defences, and discovery in the Meta case could expose pipeline practices that ripple across the industry. Expect accelerated content-licensing deals (OpenAI, Google DeepMind, and Mistral are already signing them), mandatory provenance audits, and a shift toward licensed training data. The safest move for any lab is to document data provenance and prefer licensed or one-way-crawled sources over anything peer-to-peer.
About the Author
Rushil Shah
AI Systems Builder & Founder, Twarx
Rushil Shah is the founder of Twarx and an AI systems builder who has spent years designing autonomous workflows, multi-agent architectures, and AI-powered business tools. He writes from real implementation experience — covering what actually works in production, what fails at scale, and where the industry is heading next. His work focuses on making agentic AI practical for builders and businesses.
LinkedIn · Full Profile
This article was originally published on Twarx. Follow for daily deep dives on AI agents and automation.
Top comments (0)