OpenAI opens optional text watermarking for API customers
API customers can enable textGrain for selected models. OpenAI also plans an EU rollout for ChatGPT and Codex in the coming weeks.

OpenAI announced optional text watermarking for selected API models on October 5. Customers worldwide can enable it, with the setting off by default.
The company also plans to watermark eligible ChatGPT and Codex output across EU plans over the coming weeks. Its October 5 announcement does not describe a completed rollout.
How the signal works
The method, called textGrain, changes the statistical pattern of word choices. OpenAI’s help guide says it adds no hidden characters and can be enabled through project or organization settings.
Where detection falls short
Detector applications opened October 5 for approved researchers and expert organizations. Access is restricted because watermarks can be missed and unmarked text can be flagged incorrectly.
OpenAI says short passages and tightly constrained answers are harder to detect. Editing can weaken the signal. A detected watermark does not establish authorship, ownership or accuracy.
The company previously introduced provenance tools for images and audio. The new announcement extends that work to text.
ByteForward has not independently tested textGrain.
Archival photograph of hands using a laptop by Alicia Christin Gerald, published May 31, 2026, on Unsplash under the Unsplash License. Illustrative image only. It does not show textGrain or an OpenAI product.




[…] Related coverage explains OpenAI’s optional API text watermarking. […]
[…] For a separate provenance development, see ByteForward’s coverage of OpenAI’s optional text watermarking and its detection limits. […]