Patreon Is Blocking AI Scrapers, but It's Already Too Late

Patreon just started actively blocking the AI bots that scrape creators' work to train models, and it's good to see a platform plant a flag on the creators' side. I just wouldn't mistake it for protection.

Here is what "blocking" actually means, because it's fuzzier than it sounds. For years the standard defense was a robots.txt file — a plain-text note that politely asks bots which pages to leave alone. It has no teeth, and the AI crawlers simply ignored it. Working with Cloudflare, Patreon now fingerprints the known training crawlers and refuses them at the network level, which is a real upgrade from "pretty please." In testing, one crawler's weekly hits reportedly fell from thousands to zero. Good.

The problem is scale. Patreon is a comparatively small island, and the vast libraries of creator work that AI companies actually want — YouTube, Instagram, Facebook, TikTok — belong to Google, Meta, and ByteDance, who are themselves in a full-sprint race to build frontier models. They are never going to wall off the content they already claim the right to train on. A photographer who bolts the door on their Patreon still has a decade of images sitting on Instagram, quietly feeding Meta. Locking one small platform does not matter much when the mainland is wide open and owned by the scrapers.

Then there's the timing. Blocking crawlers today does nothing about the models already trained on everything harvested up to now — the datasets are built, and you cannot un-ingest a photograph. Even going forward it's a cat-and-mouse game: crawlers rotate IP addresses, spoof their identity, and tunnel through residential proxies while the blockers chase them.

And the newest hole is the widest. Active blocking is aimed at industrial training crawlers, but a fast-growing share of AI activity now runs through agents that drive an ordinary web browser on a person's behalf — logging in, clicking, and reading a page exactly as a human would. You cannot block that without blocking your own paying members, because to the server it looks like a member. The same logic is why a paywall has never really stopped a determined AI from reading what's behind it.

None of this makes the move pointless. Precedent matters, defaults matter, and a major creator platform going on record that scraping without consent is unacceptable is worth something in the lawsuits and licensing fights still to come. But let's be honest about the size of it: this is a small company planting a flag, not a wall that keeps anything out. The training data is already gone, the biggest sources belong to the AI companies themselves, and the next wave of scraping doesn't even need a crawler — it can just log in. Patreon deserves credit for trying. I just don't think it changes where any of this is heading.