Technology
Fanfiction Community Hunts AI Writers Using Flawed Detection Tools
A new detector claims to identify Claude-generated fanfics on Archive of Our Own, but the tool risks catching innocent writers in an escalating anti-AI crackdown.
The Verge
The fanfiction world is experiencing a pitched battle over generative AI, with community members developing detection methods to root out authors they suspect of using AI tools. But the approaches being deployed are unreliable, and writers who may have done nothing wrong could pay the price.
The Claude Detector Emerges#
In late June, an anonymous X account launched a skin—essentially a browser extension—for Archive of Our Own (AO3), the massive fanfiction repository. The tool claims to identify a specific artifact: when text from Anthropic's Claude is pasted directly into AO3's editor, the chatbot wraps it with a CSS class called 'font-claude-response-body'. The skin highlights any page containing this code with a red background.
Testing appears to confirm the method works as described. When AI-generated text is pasted directly from Claude, the detector triggers; when the same text is copied through another application first, the marker disappears. But this binary signal masks deeper problems.
Why the Tool Catches Innocent Writers#
The detector can only identify Claude-generated content pasted directly into AO3—meaning any text edited elsewhere first (in Google Docs, Word, or even copied and retyped) will evade detection entirely. More significantly, the red screen cannot distinguish between a fully AI-written story and a human-written piece that simply had a few sentences run through Claude for spell-checking or translation.
One writer has already been falsely flagged after an editor used Claude to revise their work without permission. The binary nature of the detection method treats any Claude involvement as equally damning, regardless of degree.
The Anti-AI Backlash Intensifies#
Despite its creator's stated intention merely to "demonstrate" the system, fandom communities have weaponized the tool. Members are publicly naming and shaming flagged authors, treating any AI use as an inexcusable betrayal. The crackdown reflects genuine concerns: some worry about the environmental cost of AI systems and that generative models were trained by scraping content—including fanworks—from the open web.
Yet these legitimate grievances fuel an intolerant environment where detection methods remain unreliable and enforcement resembles a witch hunt.
