OpenAI to add invisible text watermarks to ChatGPT and Codex in the EU
OpenAI is adding an invisible text watermark, called textGrain, to ChatGPT and Codex output in the European Union over the coming weeks, TechCrunch and 9to5Mac report. API developers worldwide can opt in from 5 October 2026, and it is off by default. Detector access starts with approved researchers and expert organisations.
The short version
- 01OpenAI's textGrain adds a hidden statistical signal to ChatGPT and Codex text in the EU.
- 02The rollout in the EU is due over the coming weeks.
- 03API developers worldwide can opt in from 5 October 2026; it is off by default.
- 04The detector is limited at first to approved researchers and expert organisations.
- 05We label this Reported: OpenAI's own page was not available to us.
At a glance
| Name | textGrain |
|---|---|
| Where | ChatGPT and Codex in the EU |
| EU start | Over the coming weeks |
| API | Opt-in worldwide, off by default |
| Detector | Approved researchers and expert organisations |
| Announced | 5 October 2026 |
| Status | Reported |
What is OpenAI's text watermark?
TechCrunch and 9to5Mac report that OpenAI is adding textGrain, an invisible statistical watermark, to text from ChatGPT and Codex.
9to5Mac quotes OpenAI:
textGrain adds an invisible statistical signal to the model's word choices. Our detector looks for that signal to assess whether a passage contains an OpenAI watermark.
OpenAI, as quoted by 9to5Mac
Who gets the watermark?
In ChatGPT and Codex it is for users in the EU, over the coming weeks. API customers anywhere can opt in from 5 October 2026. It is off by default.
Who can check for it?
TechCrunch quotes OpenAI that the limits of the tool are why it will give initial detector access only to approved researchers and expert organisations.
Why is OpenAI doing this?
Techmeme and TechCrunch link the move to the EU AI Act, which requires AI-generated content to be marked in a way that tools can detect.
Can we confirm this on OpenAI's own page?
Not yet. OpenAI's page, Our approach to EU text provenance, did not load for us on 6 October, so we rely on two outlets that read it. That is why this story is labelled Reported.
Why does it matter?
People in the EU will be using ChatGPT text that tools can trace, even though readers see nothing different.
What's next?
The EU rollout starts in the coming weeks. We will update this story once we can read OpenAI's own page.
Questions people ask
Will I see the watermark in ChatGPT text?
No. It is invisible and works by shaping word choices.
Does it apply outside the EU?
In ChatGPT and Codex, the reports say it is for the EU. API developers worldwide can opt in.
Who can use the detector?
At first only approved researchers and expert organisations.
Has OpenAI confirmed it?
The reports quote OpenAI's announcement. We could not open OpenAI's own page, so we label this Reported.
Sources
- PrimaryOpenAI: Our approach to EU text provenance (5 Oct 2026)
- CoverageTechCrunch: OpenAI will start watermarking ChatGPT's text in the EU (5 Oct 2026)
- Coverage9to5Mac: OpenAI details new text watermarking system for ChatGPT, Codex, and the API (5 Oct 2026)
Why "Reported"? We label a story "Reported" when only the media have it so far. Spotted an error? Email hello@openloopnews.com and we will correct it.
