Breaking

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

October 5, 2026 at 14:27 EDT

Foundation Models
Software Dev & Coding
Policy & Regulation

OpenAI will add invisible textGrain watermarks to eligible ChatGPT and Codex text outputs in the European Union over the coming weeks, making the EU the first region where its consumer and coding products get default text provenance marks. The company said the rollout will apply across all ChatGPT and Codex plans in the EU, while API customers worldwide can opt in for select models and API watermarking remains off by default OpenAI announcement.

Oct. 5, 2026 | OpenAI

Invisible watermarks for EU text

Eligible ChatGPT and Codex text outputs in the European Union will receive textGrain provenance marks over the coming weeks, while API use starts as an opt-in for select models.

EU only

Default launch region

All plans

Eligible ChatGPT and Codex users

Target false-positive rate in tests

~95%

Detection for 400-token prose

Editing weakens the signal

Detection in a 400-token synonym-replacement test falls as more words change.

92%

Unedited

66%

10% edit

17%

25% edit

How the signal is used

Generate text

→

Embed word-choice pattern

→

Detector checks pattern

→

Result needs context

Why it helps

It gives platforms, publishers and compliance teams a machine-readable signal for AI-generated text provenance.

Why it is contested

Developers and researchers are watching for code effects, false positives, paraphrase removal and over-reading of detector results.

What the mark cannot prove

Identity

It does not identify a user, prompt or conversation.

Authorship

It does not measure human editing or creativity.

Accuracy

It does not verify whether the text is true.

Continue reading

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