Chatgpt watermarks

[fellow]

Anyone here good with Sublime Text? Wondering how to find hidden watermarks (like invisible characters) in AI-generated text using it. Looking for a simple method—not super experienced with coding. Thanks!

[kingkeith]

Do you have an example of AI generated text containing these watermarks? Are you sure there will be watermarks?

For invisible characters in general, thes thread may be relevant:

Is there a plugin to highlight Unicode homoglyphs (e.g., Cyrillic or Greek letters)?

Hi everyone, I’m using the Gremlins plugin to detect invisible or suspicious whitespace in Sublime Text (like non-breaking spaces, zero-width spaces, etc.), and it works well for that. However, I’ve recently run into a different kind of issue: visually deceptive Unicode characters that look like normal Latin letters but are actually from other scripts — for example: Pаypаl Verificаtiоn Reqυіred In this line: The а is Cyrillic small a (U+0430), not Latin. The υ is Greek upsilon (U+03C5). Th…

[fellow]

Hey @kingkeith

Thanks for asking — that’s actually exactly what I’m trying to learn right now using Sublime Text! I’m super curious about how AI-generated text might carry subtle “watermarks” or detectable patterns.

[DamhertPro]

I came across this site.
Maybe it will give you an idea

[alec]

I made a very simple llm markdown chat plugin that works with any openai-style API, and I added a sanitize dictionary to the settings. This automatically removes the most common weird spaces and unicode chars that ChatGPT and GPT-OSS inserts, but you can customize it if you find your favorite model inserts different characters.

The plugin has a command to sanitize the current file or selection, or you can configure it to automatically sanitize when you copy from a chat window.

GitHub - alecGraves/Agentic: Control LLM Agents from Sublime Text

[alec]

alecGraves/Agentic/blob/master/Agentic.sublime-settings#L181