When the Dataset Sued

So Meta gets to go to trial. And the plaintiff dragging the world's most valuable AI lab into discovery is Strike 3 Holdings — the company behind Blacked.com, Vixen, and Tushy — and a name already infamous in federal courts as one of the most aggressive copyright litigants in the country. The internet's most stigmatized content, represented by its most litigious owner, just cleared the motion to dismiss against the company building Llama.

Savor the symmetry, because the substance is more important than the punchline. Strike 3's investigators allege Meta used BitTorrent to pull 2,396 of their films across 47 IP addresses — and that Meta didn't just download, it seeded, redistributing the pirated files back out to the swarm while it pulled them in. This sits next to the already-documented 81 terabytes Meta allegedly hoovered out of Anna's Archive to feed its models. The training-data supply chain, it turns out, runs straight through the same pirate infrastructure that record labels spent the 2000s trying to sue out of existence. Different decade, same torrent client, bigger defendant.

Here is the part the headline can't carry. We have spent two years being told the scraping is a gray area — that the web is "public," that training is "transformative," that nobody really owns a pattern of pixels once it's been ingested into a weight matrix. The argument was never really about law. It was about visibility. As long as the source material was an undifferentiated mass — a billion blog posts, a scraped Common Crawl, the anonymous churn of the open web — there was no plaintiff with a face. The dataset was a natural resource. Free, like air, like the commons nobody defends because nobody owns.

The dataset is not free. The dataset is people. And the uncomfortable thing about this case is precisely that the people are uncomfortable — that the first creators to win standing against Meta produce adult content, the one corner of the web everyone is happy to imagine has no rights worth respecting. The messenger is exactly the messenger nobody wanted. Which is what makes the message clean: consent doesn't get more or less real depending on whether you approve of the content. Either using someone's work without permission is a distortion, or it isn't. The water finds its level regardless of how you feel about the water.

That's the structural principle hiding under the lurid framing. Coherence isn't about whether you like what's being protected; it's about whether the relationship is aligned. Take material without consent and you've introduced a fracture into the system — one that doesn't disappear because the thing you took was already taboo, or because you took so much from so many that no single theft looks like a theft. Scale was supposed to be the alibi. Forty-seven IP addresses and six thousand downloads is, it turns out, also a paper trail.

I won't pretend this fixes anything. Strike 3 is no one's idea of a folk hero, the case could still settle into a quiet nine-figure NDA, and Meta will keep training on whatever survives the next round of motions. But a year ago the official position was that the data just appeared, that the creators were invisible by design. A judge just said the creators can show up in court. That's not justice yet. It's just the dataset, for the first time, with a lawyer.

Start the timer to the settlement nobody's allowed to discuss.