A group of geeks is fighting against AI – and itself

Last week, a new group of fanworks started, with the aim of removing the authors using artificial AI. But the methods of determining what is being used are questionable, and any author of magic can be caught in the crossfire.

Big hate around using Claude, ChatGPT, and other AI tools already exist something in creative communitiesincluding the world of fiction. Readers and writers have gone through the guides to see what works they claim to have been created by AI, citing everything from line to line main idea of purple prose. But on June 29th, X’s unknown account called @heatedrivalryai he promised a seemingly reliable answer. It installed a skin – similar to an extension – of the well-known Archive of Our Own (AO3) site that would identify artifacts left behind by the Anthropic Claude bot.

“When responses created by Claude are posted directly into AO3 from Claude, the text is wrapped in Claude’s code ‘font-claude-response-body,'” the @heatedrivalryai account said. “His presence shows the use of Claude for sure.” When a user visits a page (like a fanfic service) with this code, the skin turns the entire background red.

Several exam notes published to AO3 which allows users to see if it is running. The screen turned red when I tested the skin with these samples myself, and I published a short article created by Claude to start testing myself. A red screen appeared when I wrote directly from the chatbot to the editor and disappeared if I wrote (including the same story) that did not come from Claude.

Claude’s recording was accompanied by magical examples where the objects were seen, which the anonymous developer said was designed to ensure that the system works, not “to create an environment of disbelief or to challenge other users.” But popular communities have quickly come together to name and shame authors whose published works have been recognized by the tool, and its creator does not see AI as a good thing. “Fandom is a unique, collaborative environment. It thrives on the social and creative energy that drives and feeds it,” he said. “If we unknowingly allow AI to pollute this place, what will they be left with?”

Anthropic did not respond to my request to confirm if Claude’s fan-made monitor works as described. The approach here seems clear, however, and our experiments support it. There’s no obvious reason for Claude’s code to be in the story if the bot isn’t used in some other way. But there is an obvious risk of both false negatives and excesses.

The code wrapping is saved only if the text is copied directly from Claude to the AO3 editor, so that they cannot catch anything changed in Google Docs or Microsoft Word and move it to AO3 – and as someone who writes for a living, I can testify how dangerous it is to write directly in the CMS. Some of the featured authors have already modified their works to remove the old stuff, and future works could easily escape the tool.

In contrast, the label does not reveal how much Claude was used for the task at hand. The red screen could mean that the entire story was created by an AI, or that the author pasted human-written sentences into Claude for analysis or translation, and then put them back into AO3.

This doesn’t matter to other members of the fandom, who see each one using generative AI as an unspeakable betrayal of the entire creative community. Many people cite concerns about the environmental impact of technology and how it is taught in removing the open web, which may include visuals uploaded to platforms like AO3.

The use of this tool is limited – AO3 is not the only platform for publishing preferences, and Claude is one of many AI models. At least one person says has written a separate code that can detect “Claude, Deepseek, and ChatGPT” usage, but has not released that solution to the public or explained how it works. I asked Google and OpenAI if their models would leave artifacts that could be identified in a similar way, but they did not respond.

In fact, it would be very surprising if a reliable system existed. I have been reporting story around AI awareness for several years now, and to my knowledge, there no it is currently the most reliable technical method for distinguishing between artificial and human-written documents. Systems like C2PA Content Credentials and Google’s SynthID are making strides in identifying AI output in images, videos, and even audio, but these rely on watermarks and metadata that don’t go beyond recorded text.

This may change in the future, and AI companies have every incentive to solve this problem internally. The original models were trained on words extracted randomly from the Internet, and when human text is multiplied by its artificial partners, it can be dangerous”model collapse” which may affect the accuracy of the output.

However, for now, fandom communities still rely on vibes. A lot of fun isn’t measured by a tool like the AO3 skin, but by “voices” that can include anything from unique concepts – such as the well-known “it’s not X, it’s Y” – to the overuse of floral metaphors. (At least no one in the fandom, so far, has said benches to be men.) But we must remember this AI is often spelled as such because it was taught on things written by real people. It’s trying to take us. I’m not brave enough to share my AO3 archives, but I’ve read plenty of ads in the Pre-ChatGPT internet days that wouldn’t pass this questionable test.

The best solution for differentiating AI functions on AO3 already exists: a robust site plan. A”Designed Using Generative AI” sign exists, and many authors include it to reveal the use of tools like Claude.” This requires transparency, however, there is not much incentive to act honestly because of this inconsistency.

With these efforts to prevent AI from taking a blind eye to human-driven content creation, writers who do not adhere to what is considered acceptable writing standards may be innocent of the ongoing witch hunt. At least one writer has already been caught by this because someone he trusted change their fic he did this using Claude. So if the next fanfic you’re reading feels robotic, just remember that it probably isn’t actually being a robot.