How does the Claude detector actually work?

When a user copies text directly from Claude into AO3's editor, Claude wraps the text in a code called 'font-claude-response-body.' The skin detects this code and turns the entire background red to flag the work. However, the code is only preserved if text is copied directly from Claude into AO3—if text is edited in Google Docs or Microsoft Word first and then moved to AO3, the code is not preserved and the detector will not catch it.

What are the limitations of this detection method?

The tool cannot distinguish between a fully AI-generated story and a story where an author only pasted a few sentences into Claude for spell-checking or translation. Additionally, it only detects Claude specifically; it cannot identify works using other AI models like ChatGPT or Deepseek. Some writers have already updated their flagged works to remove the artifacts, and future submissions can easily evade detection by editing text elsewhere before uploading to AO3.

Does a reliable AI detection system already exist?

According to the reporting, there is currently no reliable technological solution for distinguishing generated text from that typed out by human hands. While systems like C2PA Content Credentials, Google's SynthID, and others are making progress in detecting AI in images, videos, and audio using invisible watermarks and metadata, these methods do not carry over for copy-pasted text.

Large Language ModelsAI Coding AssistantsThe Verge AIJul 4, 2026Jul 4, 2026, 22:00 JST2 min read

Fanfiction community deploys Claude detector tool, sparking AI witch hunts

Send on LINE

Key takeaway

A fan-made detector tool for Archive of Our Own now flags fanfiction stories that contain Anthropic's Claude code artifacts, prompting fanfiction communities to publicly shame writers flagged by the tool.
However, the detector only catches text pasted directly from Claude into AO3 and cannot measure how heavily the AI was used, creating risks for innocent writers—including those whose editors used Claude without permission—while future works can easily evade the tool by editing text in other applications first.

3 Key Points

What happened
An anonymous X account released a browser skin for Archive of Our Own (AO3), a popular fanfiction repository, that flags stories containing code artifacts left by Anthropic's Claude chatbot. The tool turns the background red when it detects the Claude-injected code 'font-claude-response-body,' which is present when text is pasted directly from Claude into AO3's editor. Fanfiction communities have quickly begun using the tool to publicly identify and shame writers whose works were flagged.
Why it matters
The fanfiction community has long opposed generative AI use, citing concerns over environmental impact and AI training methods that scrape the open web—including fanworks themselves. However, the detection tool has significant limitations: it only catches text copied directly from Claude into AO3 (not text edited in Word or Google Docs first), and the red flag cannot reveal how heavily Claude was used—whether the entire story was AI-generated or just a few sentences were pasted in for spell-checking. This means innocent writers risk being caught in a witch hunt, including those whose editors used Claude without their knowledge.
What to watch
The tool's real-world applicability is limited—AO3 isn't the only fanfiction platform, Claude is just one of many AI models, and future works can easily evade detection by editing text elsewhere before posting. Notably, there is no currently reliable technological solution for distinguishing generated text from human-written text across all AI models, and it remains unclear whether other major AI providers leave similarly detectable artifacts.

Ask the AI about this article →

Summaries like this, in your inbox every morning.

Get news like this every morning — free

FAQ

How does the Claude detector actually work?: When a user copies text directly from Claude into AO3's editor, Claude wraps the text in a code called 'font-claude-response-body.' The skin detects this code and turns the entire background red to flag the work. However, the code is only preserved if text is copied directly from Claude into AO3—if text is edited in Google Docs or Microsoft Word first and then moved to AO3, the code is not preserved and the detector will not catch it.
What are the limitations of this detection method?: The tool cannot distinguish between a fully AI-generated story and a story where an author only pasted a few sentences into Claude for spell-checking or translation. Additionally, it only detects Claude specifically; it cannot identify works using other AI models like ChatGPT or Deepseek. Some writers have already updated their flagged works to remove the artifacts, and future submissions can easily evade detection by editing text elsewhere before uploading to AO3.
Does a reliable AI detection system already exist?: According to the reporting, there is currently no reliable technological solution for distinguishing generated text from that typed out by human hands. While systems like C2PA Content Credentials, Google's SynthID, and others are making progress in detecting AI in images, videos, and audio using invisible watermarks and metadata, these methods do not carry over for copy-pasted text.

The Verge AIRead Original Article

Send on LINE

Get the latest Large Language Models news every morning

For example, today's edition would include:

Visko raises $10M, launches live AI video model OrbisSiliconANGLE AI · 1h ago
Runway unveils Solaris, an AI that generates app interfaces in real timeTHE DECODER · 1h ago
Google AI Search flags Facebook users as dangerTHE DECODER · 1h ago

AI-summarized, only the topics you pick — one digest a day via Email, Slack, or Discord.

Subscribe free →

Free · takes 30 seconds · unsubscribe anytimeWhat is AIToday? →