A red screen. That's what greets you now on certain fanfic pages. Not a review. Not a kudos. A scarlet flag planted by a browser skin that claims to smell artificial intelligence.

The skin, posted June 29 by an anonymous X account called @heatedrivalryai, hunts for a single string: font-claude-response-body. Anthropic's Claude wraps every response in that class. Paste directly from the chat window into Archive of Our Own's editor, and the wrapper follows. The skin sees it. The background bleeds red.

I tested it. The red screen appears on direct paste. It vanishes if you copy the same text into Notepad first. It vanishes if you write the prompt in ChatGPT. It vanishes if you edit a single word.

This is not a detection system. This is a trap for the lazy.

Yet fandom didn't pause to read the fine print. Within hours, screenshots circulated. Lists formed. Names were called out in Discord servers and subtweeted on timelines. The detector's creator insisted they didn't want "an environment of mistrust." They got a witch hunt anyway.

The contract is broken

Fanfiction runs on a gift economy. You write. I read. We trust the human at the other end of the keyboard. That trust is the whole infrastructure — no DRM, no paywalls, no verification badges. Generative AI breaks the contract. It lets someone harvest the corpus, remix it at scale, and pass the output off as labor.

Writers have every right to feel violated. Readers have every right to feel cheated. The revulsion is earned.

But the response has been performative purity theater. The Claude detector catches one model, one workflow, one failure mode. It misses Gemini. It misses GPT-4o. It misses the writer who prompts, curates, rewrites, and claims authorship. It misses the writer who uses AI for outlines only. It misses the writer who used AI once, three years ago, for a single paragraph they've since rewritten twice.

All of them walk free. Only the copy-paster burns.

False confidence is worse than no confidence

The methodology holds up, technically. Anthropic didn't respond to verification requests, but the artifact is real. The code exists. If the wrapper appears, Claude touched the text. That much is solid.

The leap from "Claude touched this" to "this author is a fraud" is where the rot sets in.

Consider what the wrapper actually proves. It proves the author opened Claude, generated something, and pasted it without stripping metadata. That's it. It doesn't prove the work is AI-written. It doesn't prove the author didn't write 90% of it themselves. It doesn't prove they didn't use the tool to translate their own draft from a language they speak poorly into English they speak worse.

Context is everything. The detector strips it all away.

The shaming machine

Fandom has always policed its own boundaries. Plagiarism callouts. Tag policing. Discourse over tropes and ethics. But those fights played out in comments, in private messages, in moderated communities with rules and receipts. The AI panic has no moderation. No due process. No appeal.

A red screen is a sentence. The mob executes it.

Writers flagged by the tool have deleted accounts. Some posted defenses — screenshots of their draft history, timestamps, the Notepad intermediary that stripped the wrapper. The defenses don't matter. The accusation traveled faster. The stain remains.

This is not community defense. This is community cannibalism.

The real threat wears a different face

The sophisticated AI user — the one prompting in chains, feeding character bibles, iterating through twenty drafts, weaving the output into their own voice — will never trigger this detector. They don't need to paste from the chat window. They use APIs. They use local models. They use tools that leave no fingerprints.

The detector catches amateurs. The professionals sail through.

And the professionals are the ones flooding Kindle Unlimited, Substack, and yes, AO3 with volume that buries human work. They don't care about fandom's honor system. They care about output per hour. No browser skin stops them.

No technical fix for a social fracture

AO3's parent organization, the OTW, has stayed silent. They've historically resisted automated moderation, citing the impossibility of fair enforcement at scale. They're right. Any detector deployed platform-wide becomes a weapon. False positives destroy reputations. False negatives breed complacency. The arms race favors the better-funded side — the AI companies, not the volunteer coders.

The solution, if one exists, isn't code. It's culture.

Fandom needs to decide what it values more: the purity of the process or the humanity of the participants. Right now, it's choosing the spectacle of the purge. Red screens are easy. Trust is hard. Rebuilding trust after you've burned your neighbors for a wrapper class they may not have even known existed? That's the work no skin can automate.

The war against AI was never winnable this way. The war against each other? That one's just getting started.