textGrain
OpenAI will add invisible textGrain watermarks to ChatGPT and Codex text for EU users on all plans, with an API opt-in worldwide.
textGrain embeds a statistical signal in the model's word choices. A secret key decides which word is picked where several would fit equally well, so no hidden characters are inserted. The rollout to EU users happens over the coming weeks. API customers worldwide can opt in for select models, and the setting is off by default. The detector is limited to approved researchers and expert organisations.
- Date
- Monday 5 October 2026
- Lab
- OpenAI
- Kind
- feature
- Access
- app only
Figures
| Measure | Value | Measured by |
|---|---|---|
| Detection rate at 200 tokens | about 80% | company |
| Detection rate at 400 tokens | about 95% | company |
| Detection after replacing 10% of words | 66% down from 92% | company |
| Detection after replacing 25% of words | 17% | company |
| False positive target | 1% | company |
Responds to Article 50 of the EU AI Act, with a compliance deadline of 2026-12-02 for existing providers. The API opt-in is not limited to the EU.
Sources
This record was partly confirmed: some claims could not be checked on 8 October 2026. How we check