Rogue Employees Don't Download 81,000 Videos

There's a sentence in this ruling that should make every AI company's legal team go very quiet.

A federal judge has found that Meta's attempt to attribute the scraping of thousands of adult videos — from sites including Vixen.com and Tushy.com — to rogue employees "strains credulity." That's the phrase. Strains credulity. From a judge. In a written ruling. About one of the most valuable companies on Earth.

A writer at 404 Media covered the story straight: Blacked.com's parent company can proceed with its lawsuit against Meta over the scraping of its content for AI training purposes. The rogue-employee defense didn't hold. The case moves forward.

What interests me isn't the porn. It's the defense that failed.

The Defense Was Always the Tell

The rogue-employee framing is a particular kind of legal maneuver — one that only makes sense if you're trying to create distance between a corporation and something the corporation clearly did. It's the same logic as a bank blaming a teller for a policy. It works when the act is genuinely isolated. It falls apart the moment scale enters the picture.

Thousands of videos don't download themselves through individual bad actors operating outside company knowledge. A judge apparently agrees. And once a court decides that explanation strains credulity, you're no longer arguing about what happened. You're arguing about what to do about what happened.

That's a different conversation. A much harder one.

The AI industry has spent the better part of three years developing a sophisticated vocabulary for not answering the consent question directly. Training data is discussed in aggregate. Sources are described as "publicly available." The humans who made the underlying content are present in the output and absent from the credit. This has worked — not because anyone was fooled, but because the legal infrastructure to challenge it was still being assembled.

That infrastructure is now being assembled.

What Survives Honesty

Here's the version of this story that nobody wants to tell: the content that made large language models and image generators genuinely impressive came from somewhere. Writers wrote it. Photographers shot it. Filmmakers filmed it. And yes, adult content studios produced it, which turns out to carry the same copyright protections as anything else.

The industry bet that moving fast enough would make the sourcing question academic. By the time anyone organized a legal challenge, the models would be too embedded, too useful, too profitable to meaningfully unwind. Maybe that bet still pays off. But a judge ruling that a major tech company's explanation for how it acquired training data "strains credulity" is not a small moment. It's the first time the gap between what the industry says it does and what it actually does has been named, plainly, by someone with authority to name it.

The companies that figure out how to be honest about their data sourcing — before a courtroom forces the honesty — are going to be in a different position than the ones who kept running the same defense until it stopped working.

Creativity doesn't need a rogue employee. It needs a policy.