OpenAI will watermark ChatGPT's writing in the EU and publishes where the watermark fails
OpenAI says it will hide an invisible signal, which it calls textGrain, in text that ChatGPT and Codex write for people in the European Union. A watermark here means a marker that software can detect but readers cannot see.
Why it matters
The rollout covers EU users only. The useful part for everyone else is the set of limits OpenAI published itself. Replacing 25% of the words with synonyms cut detection to 17%, and a watermark cannot show who wrote a passage. Keep that in mind if a client asks whether your copy was AI-written.
Your moveRead the limitations section of OpenAI's announcement before trusting any AI-text detector result.
Read the source