Regulation
ChatGPT and Codex will watermark their text in the EU, but editing it weakens the mark
OpenAI will add an invisible signature to ChatGPT and Codex text in the European Union only. A detector can read it, but for now only approved researchers will use it.
The essentials
2 confirmed facts · 2 according to the source · 3 open questions
- OpenAI will mark ChatGPT and Codex text with an invisible signature in the European Union only, on all plans, in the coming weeks.13
- The move responds to the transparency rules of the EU AI act, in force since 2 August.23
- According to the sourceIn an OpenAI test, replacing 10% of the words with synonyms cut detection from around 92% to 66%.3
- According to the sourceFor now, the detector will only be available to researchers and expert organizations approved by OpenAI.3
Why it matters
EU law requires AI-generated content to be marked in a way that other systems can recognize, and OpenAI responds with a mark that can be weakened by editing the text. If what the company says holds up, the system works mainly as a clue, not as proof. OpenAI warns that not finding the mark does not show that a text is human-written: it may be very short, heavily edited or come from another AI3.
If you use ChatGPT or Codex from the EU, your text will carry the mark without you noticing. OpenAI says it does not identify you, and for now nobody outside the approved researchers will be able to check it.
The details
OpenAI announced on 5 October 2026 that it will start marking ChatGPT and Codex text to comply with the EU AI act, according to TechCrunch. The mark will reach EU users on all plans in the coming weeks. Developers who use its API, in any country, can turn it on now for some models, but it is off by default. OpenAI is not making it a global standard for now3.
The mark is not a symbol. The model slightly adjusts which words it picks, and that leaves a pattern the eye cannot see but a detector can. Because it is in the words themselves, it travels with the text when you copy and paste it. OpenAI says it does not identify the person writing and that it saw no significant changes in the performance of its models3. Its method is called textGrain and was published with a technical report3.
What we don't know
- Whether OpenAI will take the mark outside the EU or turn it on by default in other countries.
- When the general public, and not just approved researchers, will be able to use the detector.
- Whether independent third parties will confirm the reliability that OpenAI claims for its method.
Background
Anthropic announced two months ago that it would mark Claude's text, and it applies this worldwide. Some users protested because they consider themselves the authors of the text and see Claude only as a tool, according to TechCrunch3. OpenAI had already built a text watermark and left it unreleased, partly out of fear that users would switch to rivals without one, according to The Wall Street Journal as cited by TechCrunch3.
Sources
- 1Our approach to EU text provenance rules
- 2OpenAI is adding text watermarking in ChatGPT and Codex
- 3OpenAI will start watermarking ChatGPT’s text in the EU
Was this story useful? Yes Not really
This story was written in Spanish by an artificial intelligence system from the sources listed above, and translated automatically. Before publication, a program checks that every figure and every name appears in the sources, and that the translation keeps the same facts, figures and sources. It has not been reviewed by a person. Spanish original · How we work (in Spanish)