Someone created a browser extension in June 2026 that can detect text written by Claude, an AI chatbot made by Anthropic. The tool works on Archive of Our Own, the world's largest fanfiction website. When it finds Claude-written text, it turns the page background red.

Here's how it works. When you copy text directly from Claude and paste it into AO3's editor, Claude leaves behind a hidden marker in the code — like a fingerprint. The extension, created by an anonymous account called @heatedrivalryai, scans for that fingerprint. Find it, and the page turns red. The Verge tested it and confirmed it works.

But the tool has a weakness: it only catches text pasted directly from Claude. If you copy the text into Google Docs or Microsoft Word first, then paste it into AO3, the fingerprint disappears. The tool catches people who copy-paste without extra steps, not necessarily people trying to hide AI writing.

The person who created the extension said the goal was simple: show that detection is possible. They did not want to shame anyone. But that is not what happened. Fans started publicly calling out writers whose pages turned red, and some of those writers quietly deleted or edited their posts. The tool became a way to catch and shame people much faster than the creator expected.

AO3 already has a system for fans to label their own work. You can tag a story as 'AI-Generated Text' or 'Created Using Generative AI' if you want to be honest about it. The tag system exists, but it relies on people choosing to use it. The question the fanfiction community keeps asking is: what do we do when people do not label their work voluntarily?

Anthhropic, the company that made Claude, did not answer questions about whether the fingerprint the tool detects is real or whether it stays the same as Claude gets updated.

That silence is important. The fingerprint the tool looks for is not a permanent part of Claude. It is just a detail about how Claude is currently built. Anthropic could change it, remove it, or scramble it any time with a software update. If someone relies on this tool to catch AI writing for real enforcement, they could be fooled next month when Claude changes. The tool works right now, but it might not work forever.

What this really shows is something bigger. Fanfiction communities have a trust problem that no detection tool can solve by itself. Fanfiction is built on the idea that fans are creating their own stories — putting real thought and heart into them. When fans object to AI-written fanfic, it is not really about whether it is good or bad. It is about whether the person writing it did their own creative work. A red page proves someone pasted from Claude. It does not solve what the fanfiction community actually wants to decide: where to draw the line between human-written and machine-written stories.