The irony is so thick you could cut it with a GPU.
For years, the big AI players—Anthropic, OpenAI, Google—have been basically saying: “If it’s on the internet, it’s fair game for our training data.” They’ve scraped websites, hoovered up content, and called it “fair use.” Website owners complained. Lawyers got involved. Nothing changed.
Now? These same companies are absolutely *losing it* because competitors are doing the exact same thing to them.
Welcome to “distillation”—the practice of using one AI model’s outputs to improve another. Anthropic is particularly upset, claiming rivals are harvesting its model outputs at scale to shortcut billions of dollars in research. OpenAI and Google are making similar complaints. The fear is real: why spend billions building the best AI if someone else can just copy your homework for pennies?
Fair point. But here’s where it gets delicious.
## The Symmetry Problem
From a bird’s-eye view, distillation looks *exactly* like what these companies have been doing to the rest of the internet. Grab content for free. Don’t ask permission. Turn it into a product you sell. Claim it’s fair use. Hope the lawyers figure it out later.
Website owners have been screaming about this for three years. Now AI companies are screaming about the same thing. The symmetry is almost poetic.
And get this: Anthropic—the company that literally markets itself as the *ethical* AI company—is apparently the worst offender. Its data-sucking bots crawl websites thousands of times for every single referral they send back. Thousands to one. That’s not just aggressive; that’s contemptuous.
## The Bot Wars
Anthropic, OpenAI, and Google frame this as a cybersecurity issue—swarms of bots “attacking” their models. But they’ve been doing the same thing to websites, bombarding them with so much bot traffic that site owners’ cloud bills have skyrocketed. So not only are websites having their content stolen, they’re *paying more* for the privilege.
The AI industry can’t even agree on what distillation is or whether it’s okay. There’s the “benign” version (using your own model outputs to create smaller models) and the “attack” version (using someone else’s outputs). But the lines blur constantly, and now researchers are worried that Anthropic’s aggressive stance will kill all types of distillation—even the legitimate kind.
## The Inevitable Conclusion
Here’s the thing about the modern internet: once information goes online, clever people will find ways to use it. That’s true for blogs, photos, code, videos, and yes, AI model outputs. It’s always been a cat-and-mouse game, and it always will be.
Anthropic has spent months tightening access to its top models to stop competitors from learning too much. It either backfired or just inspired more creative workarounds. Because that’s how the internet works.
Distilling another company’s AI model might even be fair use. These legal arguments cut both ways.
So here’s the message for Anthropic, OpenAI, and Google: Welcome to the internet. Get used to it.
