Breaking

OpenAI Will Add textGrain Watermarks to ChatGPT and Codex Text in the EU

October 5, 2026 at 11:24 EDT

Policy & Regulation
Security
Foundation Models

OpenAI said it will add invisible text watermarks to eligible ChatGPT and Codex text output in the European Union over the coming weeks, while letting API customers worldwide opt in for select models. The company is calling the watermarking system textGrain and says detector access will begin with approved researchers and expert organizations, not the general public.

Oct. 5, 2026 · OpenAI

EU text watermarks arrive

OpenAI will add invisible textGrain signals to eligible ChatGPT and Codex text in the EU, while API customers worldwide can opt in and detector access starts with researchers.

false-positive target

95%

400-token psychology detection

66%

after 10% synonym edits

17%

after 25% synonym edits

Editing erodes the signal

In a 400-token test, synonym replacement cut detection sharply. Column height is proportional to the reported detection rate.

92%

Unedited

66%

10% replaced

17%

25% replaced

How the check works

Token choices

The model has multiple plausible next words.

→

Keyed signal

textGrain biases choices with hidden statistics.

→

Detector result

Approved reviewers test for an OpenAI watermark.

Compliance path

EU rollout for eligible ChatGPT and Codex users, plus opt-in watermarking for API customers worldwide.

Limits remain

Short, constrained, edited or translated text can weaken detection; a mark does not prove authorship or accuracy.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.