The fanfiction community is at war with AI and itself

Over the past week, a new fanwork movement has taken off, aiming to root out authors who use generative AI. But the detection methods carried out are questionable, and any fiction writer could be caught in the crossfire.

Widespread aversion There has been a long discussion about the use of Claude, ChatGPT, and other AI tools something in Creative communitiesincluding the world of fan fiction. Readers and writers have passed along tips for spotting works supposedly created by AI, citing anything from dashes to… Broad concept From purple prose. But on June 29, an anonymous X account called @heatedrivalryai Promising a seemingly more reliable solution. I’ve posted a skin – resembling an extension – of the popular Archive of Our Own (AO3) fan archive repository that purports to identify elements of programming left behind by Anthropic’s Claude bot.

“When the response generated by Claude is pasted directly into Claude’s AO3, the text is wrapped by code injected by Claude ‘font-clude-response-body,’” account @heatedrivalryai said. “Its presence indicates Claude’s ultimate use.” When a user visits a page (such as a fan art) using this icon, the skin turns the entire background red.

numerous Test posts It is published on AO3 which allows users to check whether it is working or not. The screen immediately turned red when I tested the skin against these examples myself, and I posted a short story written by Claude to run my own experiment just in case. The red screen appeared when I pasted directly from the chatbot into the editor and disappeared if I pasted text (including the same generated story) and it didn’t come directly from Claude.

Kashif Claude’s post was accompanied by fictional examples where the artifacts had been spotted, which the anonymous creator said was intended to demonstrate the working of the system, not to “create an environment of mistrust or accuse certain users.” But fan communities quickly mobilized to name and publicly shame writers whose published work was flagged by the tool, and its creator certainly doesn’t view AI as a positive thing. “Fandom is a uniquely connected and collaborative space. It thrives on the human element and the creative spark that drives and fuels it,” they said. “If we inadvertently allow AI to corrupt these spaces, what will be left?”

Anthropic did not respond to my request to check if the fan-made Claude detector worked as described. However, the methodology here seems sound, and our tests support it. There’s no obvious reason for Codecloud to be in the story if the robot isn’t being used in some way. But there is a clear risk of false negatives and overgeneralization.

Code wrapping is only preserved if the text is copied directly from Claude into the AO3 editor, so anything edited in Google Docs or Microsoft Word and then transferred to AO3 won’t be captured – and as someone who writes for a living, I can attest to how dangerous it is to write directly into a CMS. Some flagged writers have already updated their works to remove artifacts, and future works can easily evade the tool.

On the contrary, the mark does not reveal how intensively Claude is used in a particular work. That bright crimson screen could mean that the entire story was entirely AI generated, or that the author pasted some human-written sentences into Cloud for spell-checking or translation, and then sent them back to AO3.

This didn’t matter to some members of the fanbase watching any The use of generative AI as an unforgivable betrayal of the broader creative community. Many people point to concerns about the environmental impact of the technology and how it can be trained by scraping the open web, which likely includes fanworks uploaded to platforms like AO3.

The applicability of this particular tool is limited — AO3 is not the only platform for publishing fanworks, and Claude is just one of many AI models. at least One person claims They’ve written separate code that can detect use of “Claude, Deepseek, and some ChatGPT,” but haven’t released that solution publicly or explained how it works. I asked Google and OpenAI whether their models leave any traceable traces in text generation that could be detected by similar means, but they did not answer.

In fact, it would be very surprising if there was a universally trusted system. It has been reported Issues surrounding Artificial intelligence detection For a few years now, and as far as I know, there is not so It is currently a reliable technological solution to distinguish generated text from text written by human hands. Systems like C2PA Content Credentials and Google’s SynthID are making some progress toward identifying generative AI in images, videos, and even audio, but they rely on watermarks and invisible metadata that doesn’t carry over for copy-and-paste text.

This may change in the future, and AI companies have every incentive to at least solve the problem internally. Early models were trained on text randomly extracted from the Internet, and with human writing crowding out by its synthetic counterpart, they could risk “Model collapse“A scenario that would reduce the accuracy of the output.

Currently, mass societies still rely mostly on emotions. Most fans are judged not by a tool like the AO3 theme, but by “telling” which can include anything from specific sentence structures – like the infamous “It’s not X, it’s Y” – to the overuse of flowery metaphors. (At least no one in the fandom has yet Benches become men.) But we have to remember that AI is often written this way Because it was trained on stuff written by real people. She is trying to imitate us. I’m not bold enough to share my AO3 mentions, but I’ve certainly read some overly eloquent fiction in the pre-ChatGPT internet days that wouldn’t pass this dubious sniff test.

The best solution for highlighting AI work on AO3 is already available: the site’s robust tagging system. A “Created using generative artificial intelligence“The tag exists, and many authors include it to disclose the use of tools like Cloud. However, this requires honest transparency, and there is little incentive for honesty given the backlash. It is also worth noting that fanfiction is meant to be a hobby, not a regulated industry.

With these efforts to prevent AI from taking attention away from true human-driven creativity, authors who do not conform to what is considered acceptable quality of writing may become innocent victims of the ongoing witch hunt. At least one writer has actually fallen into this because of someone else they trust To edit their photos I did this using Claude. So, if the next fanfiction you read sounds like it’s a bit robotic, just keep in mind that it may not be In reality It is the product of a robot.