Breaking

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

October 6, 2026 at 04:40 EDT

Policy & Regulation
Foundation Models

OpenAI will begin adding an invisible textGrain watermark to eligible ChatGPT and Codex text output in the European Union, making its text provenance program a live response to the EU AI Act rather than a research promise. The company said API customers worldwide can opt in to watermarking for select models now, while consumer ChatGPT and Codex rollout in the EU will arrive over the coming weeks across all plans.

October 2026 · OpenAI

Invisible watermarks reach ChatGPT text

OpenAI is adding textGrain to eligible ChatGPT and Codex text in the EU, while API customers can opt in worldwide for select models.

target false-positive rate

80%

200-token passage detection

95%

400-token passage detection

200

token threshold for short text

Editing weakens the signal

Detection on 400-token passages fell sharply as more words were replaced with synonyms.

92%

no synonym swap

66%

10% synonyms

17%

25% synonyms

How the rollout works

API opt-in
Global customers can enable select models.

→

EU default
Eligible ChatGPT and Codex text is marked.

→

Limited detector
Access starts with approved experts.

What it helps

Adds a machine-readable provenance signal for AI-generated text, supporting compliance, review, and content authenticity workflows.

What it cannot prove

It does not prove authorship, ownership, accuracy, or user identity, and rewriting or short text can make detection unreliable.

Continue reading

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