Will ChatGPT watermark my text if I am outside the EU?

OpenAI's ChatGPT and Codex watermarking applies only to eligible text generated in the European Union, across all plans, and rolls out over the coming weeks. OpenAI says textGrain will not become a global default at launch. Outside the EU, the watermark only appears when an API customer chooses to turn it on for select models.

Can anyone check whether a text has an OpenAI watermark?

Not yet. OpenAI limits access to the textGrain detector to approved researchers and expert organizations, granted case by case, because short passages and later edits make detection less reliable. The detector needs only the text and OpenAI's secret key. OpenAI says it plans to release the technology as open source so others can build on it.

How easy is it to remove the textGrain watermark?

Editing weakens OpenAI's textGrain signal quickly. In OpenAI's 400-token tests, replacing 10% of words with synonyms cut detection from about 92% to 66%, and replacing 25% of words cut it to 17%. Detection is also much lower for math content, where the model has less freedom in word choice.

Does watermarking make OpenAI model output worse?

OpenAI says that across the benchmarks it uses to assess Astra, which it calls its latest frontier model, it sees no meaningful performance difference with and without textGrain watermarking. The method caps how much sampling entropy the watermark removes, which is meant to keep output quality close to unwatermarked text.

What can an OpenAI text watermark prove?

An OpenAI textGrain watermark can show that an OpenAI system generated or processed part of a text. OpenAI says it cannot show who used the system, how much a person contributed, who owns the text, or whether the content is accurate. If no watermark is found, that also does not prove a person wrote the text.

OpenAI · 2026-10-05 · major

OpenAI textGrain — invisible text watermarks for the EU AI Act

OpenAI now lets API customers worldwide opt in to textGrain text watermarking, off by default. Over the coming weeks it adds the invisible watermark to eligible ChatGPT and Codex text made in the EU, to meet AI Act Article 50.

OpenAI Developer Community banner on a blue gradient background

OpenAI starts marking AI-written text in the EU and lets any API customer turn the same watermark on.

Quick facts

Maker	OpenAI
Technology	textGrain statistical text watermark
API	Opt-in worldwide for select models, off by default
ChatGPT and Codex	EU only, all plans, rolling out over coming weeks
Detection (1% false positives)	~80% at 200 tokens, ~95% at 400 tokens
Detector access	Approved researchers and expert organizations, case by case
Legal driver	EU AI Act Article 50 (applies from 2 Aug 2026)

What is it?

textGrain is OpenAI's new invisible watermark for model-written text. From 5 October 2026, API customers in any country can switch it on for select models; it stays off by default. Eligible ChatGPT and Codex text generated in the European Union gets the watermark over the coming weeks, on all plans, but it is not a global default.

How does it work?

The watermark hides a statistical signal in the words the model picks: generation is tied to keyed randomness, with an entropy budget that limits how much sampling freedom the signal takes away. A detector needs only the text and OpenAI's secret key to test for the signal. At a 1% false-positive target, OpenAI reports about 80% detection on 200-token passages and about 95% on 400-token passages, with much lower rates for math, where word choice is less flexible.

Why does it matter?

Article 50 of the EU AI Act requires generative AI output to be machine-detectable, and its transparency rules apply from 2 August 2026. Developers who ship OpenAI-generated text to EU users now have a provider-side switch for that duty. The limits are real: replacing 25% of words with synonyms cuts detection to 17%, and OpenAI says a missing watermark does not prove a person wrote the text.

Who is it for?

API developers serving EU users, compliance teams, researchers studying AI-text detection

Frequently asked questions

Will ChatGPT watermark my text if I am outside the EU?: OpenAI's ChatGPT and Codex watermarking applies only to eligible text generated in the European Union, across all plans, and rolls out over the coming weeks. OpenAI says textGrain will not become a global default at launch. Outside the EU, the watermark only appears when an API customer chooses to turn it on for select models.
Can anyone check whether a text has an OpenAI watermark?: Not yet. OpenAI limits access to the textGrain detector to approved researchers and expert organizations, granted case by case, because short passages and later edits make detection less reliable. The detector needs only the text and OpenAI's secret key. OpenAI says it plans to release the technology as open source so others can build on it.
How easy is it to remove the textGrain watermark?: Editing weakens OpenAI's textGrain signal quickly. In OpenAI's 400-token tests, replacing 10% of words with synonyms cut detection from about 92% to 66%, and replacing 25% of words cut it to 17%. Detection is also much lower for math content, where the model has less freedom in word choice.
Does watermarking make OpenAI model output worse?: OpenAI says that across the benchmarks it uses to assess Astra, which it calls its latest frontier model, it sees no meaningful performance difference with and without textGrain watermarking. The method caps how much sampling entropy the watermark removes, which is meant to keep output quality close to unwatermarked text.
What can an OpenAI text watermark prove?: An OpenAI textGrain watermark can show that an OpenAI system generated or processed part of a text. OpenAI says it cannot show who used the system, how much a person contributed, who owns the text, or whether the content is accurate. If no watermark is found, that also does not prove a person wrote the text.