Fanfiction found an AI detector that raises alarms faster than it proves authorship

The Verge reports a new conflict inside fanfiction communities: readers are trying to identify work written with generative AI, and some are using an AO3 skin aimed at Claude. The tool can find a specific technical artifact, but the community is turning that artifact into a broader judgment about authorship.

The AO3 skin looks for a direct paste trail from Claude

According to The Verge, an anonymous account called @heatedrivalryai published the skin for Archive of Our Own on June 29. It is meant to flag text that carries the font-claude-response-body class after being pasted directly from Claude into AO3.

The Verge’s tests found that the page turned red on examples carrying that trail. The same generated text was no longer flagged when it did not come straight from the chatbot into AO3. Anthropic did not respond to the publication’s request to verify the behavior.

The AI fight has moved from prose style to workflow forensics

The broader point goes beyond fanfiction. AI detection often gets framed as a matter of style, but this case is not about literary taste. It is about formatting residue left by a publishing workflow.

That matters for creative platforms. When a community lacks trusted AI labeling rules, it starts building its own police force. This approach is technically smarter than guessing from long sentences, but it still sees only one narrow signal.

A red screen does not say who wrote the story

The limitation is fundamental. The artifact survives only on a specific copy path. A story edited through Google Docs or Microsoft Word may not carry it even if AI was used heavily. A flagged story also may not be fully generated: the author might have run a few human-written lines through Claude for translation or proofreading.

That is where a technical signal becomes a social risk. Publicly shaming an author based on one CSS class is a weak evidentiary standard, even if the artifact itself is real.

Platforms need rules better than artifact hunting

The next signal is how platforms respond. If AO3 and similar services leave enforcement to community skins and viral lists, suspicion becomes cheaper than proof.

The better path is explicit policy: what counts as unacceptable AI use, when disclosure is enough, how appeals work and how writers are protected from false accusations. Without that, every new detector becomes a match, not a fire extinguisher.

Lilith's verdict

The fandom built a red alarm and started using it like a judge’s gavel. The problem is that the alarm watches a copy trail, while the person in the dock is an author with a reputation to lose.