The new ChatGPT models leave extra characters in the text — they can be «detected» through Word

The new ChatGPT models leave extra characters in the text — they can be «detected» through Word

From now on, ChatGPT will be harder to hide. It turns out that with the release of the new o3 and o4-mini models, OpenAI chatbot started leaving hidde

itc.ua

TL;DR: OpenAI’s latest o3 and o4-mini ChatGPT models are sneakily dropping invisible watermarks—special Unicode “spaces” that look like normal blanks but pop up as circles if you toggle hidden characters in Word (Ctrl + Shift + 8) or a code editor. OpenAI insists it’s a bug, not a feature, and older GPT releases are watermark-free.

Cool hack: Just open the text in any editor that shows special chars, search-and-replace the funky spaces with regular ones, and voilà—your watermark-free ChatGPT output.