A Browser Skin That Turns AO3 Red: Fanfic Communities Build Their Own Claude Detector

An anonymous X account called @heatedrivalryai posted a browser extension on June 29, 2026 that detects Anthropic's Claude-generated text on Archive of Our Own — and the fanfiction community has since mobilized around it in ways its creator did not entirely anticipate.

The mechanism is specific. When a user pastes text directly from Claude into AO3's editor, the AI leaves behind a residual CSS class: font-claude-response-body. The @heatedrivalryai extension — framed as a "skin" in AO3 parlance — scans for that wrapper on any AO3 work page. If it finds it, the entire page background turns red. The Verge tested the tool and confirmed the behavior: red screen on direct Claude-paste, clean background on the same story stripped of the artifact. Several test works were published on AO3, including at works/87682756, for community verification.

The limitation is narrow but exploitable. The detection fires only on text copied straight from Claude into AO3's editor. Route the same output through Google Docs or Microsoft Word first and the font-claude-response-body wrapper does not survive. The tool, in other words, catches laziness more than intent — a distinction worth holding onto.

The account's creator has been explicit about the original purpose: to demonstrate that detection is technically possible, not to "create an environment of mistrust or accuse particular users." That framing has not held. Community members have already begun publicly naming and shaming writers whose works triggered the red screen, and some of those writers have quietly updated their AO3 posts to scrub the Claude artifacts. The tool became a social instrument faster than its author anticipated.

AO3 already has voluntary disclosure infrastructure. The platform maintains browsable tag pages for both 'AI-Generated Text' and 'Created Using Generative AI', and a separate tag exists for 'AI-Generated Images' — so the architecture for self-labeling has been there. The question the community has consistently wrestled with is what happens when that labeling is not voluntary. The Organization for Transformative Works addressed the broader tension back in May 2023, when its OTW Signal newsletter and a dedicated page on AI and data scraping both acknowledged community anxiety and referenced existing detection tools including ZeroGPT and GPTZero. The Claude skin is a more targeted instrument than those probabilistic classifiers, but it sits in the same contested space.

Anthropic did not respond to The Verge's request to verify whether the artifact the tool detects is genuine or whether it persists across Claude versions.

That silence is notable for a practical reason. CSS class names injected by a web-based AI interface are an implementation detail, not a deliberate provenance signal. They can be renamed, removed, or randomized in any product update. The current detector works against a specific Claude behavior at a specific moment; it is not a durable forensic standard. Anyone relying on it for enforcement rather than demonstration is building on a foundation Anthropic could dissolve without announcement.

Looking at what this actually exposes: the fanfic ecosystem is dealing with a trust problem that AI detection tooling — probabilistic or artifact-based — cannot cleanly resolve. The AO3 community is built on a shared understanding that the works are expressions of human creative labor, often deeply personal, always relational to a source text and to other fans. The objection to AI-generated fic is not primarily about quality; it is about that relational contract. A red screen proves a particular paste operation happened. It does not settle what the community actually wants to negotiate, which is the boundary between human authorship and machine generation in a space that has always valued the former as a given.

The broader pattern is familiar from other content platforms — the cycle runs from voluntary disclosure norms, through community-built detection tools, toward eventual platform-level policy. AO3 and the OTW have been characteristically deliberate; no enforcement mechanism exists beyond social pressure and self-tagging. Whether the Claude skin accelerates a harder policy conversation or simply becomes an arms-race footnote depends largely on whether Anthropic's next interface update retains the artifact. If it does not, the red screen goes dark and the underlying tension remains exactly where it was.