OpenAI will add an invisible watermark to ChatGPT text in the EU
The change follows the EU AI Act’s transparency rules and rolls out over the coming weeks. OpenAI says its detector will go first to approved researchers, not the public.
How this story was checked
- ResearchedWith AI assistance · 3 sources
- AI source checkEach sentence matched to its source
- Editor approvalMayank Sahu · 11 Oct, 20:32
- Published11 Oct, 20:32
In short
OpenAI said on 5 October 2026 that ChatGPT and Codex text for users in the EU will carry an invisible watermark, rolling out over the following weeks on all plans. API users worldwide can opt in. The detector is limited to approved researchers at first.
What is changing
OpenAI said on 5 October 2026 that text written by ChatGPT and Codex will carry an invisible watermark for users in the European Union, on all plans, rolling out over the following weeks.1 It is not a global default at launch.12
API customers anywhere can turn watermarking on for selected models from the same day. It stays off by default in the API.1
Why now
TechCrunch reports that OpenAI says the change is needed to comply with the EU AI Act’s transparency rules, which require AI-generated content to be marked so that other systems can detect it.2 The European Commission says these transparency requirements apply from 2 August 2026.3
How the watermark works
OpenAI calls the method textGrain. It slightly shapes the model’s word choices, leaving a statistical pattern that a reader cannot see but a detector with a secret key can find.12 Because the signal is in the words, it survives copy and paste.2 OpenAI says the watermark does not identify the user, and reports no meaningful change in output quality.1
How well it works
OpenAI’s own test figures, at a 1% false-positive target:1
- About 80% detection on 200-token passages and about 95% on 400-token passages.
- Replacing 10% of words with synonyms cut detection on 400-token passages from about 92% to 66%. Replacing 25% cut it to 17%.
- Detection is much weaker for maths answers, and short or translated text is harder to detect.12
Who can check text
The detector is not public. OpenAI is taking applications from researchers and expert organisations and granting access case by case.1 The detector reports only whether an OpenAI watermark is present. It does not identify the user or reveal prompts.1
What a watermark does not prove
OpenAI says a missing watermark does not prove that a human wrote something, and a watermark does not show how much human effort went into a passage.1 TechCrunch notes that Anthropic said in August it would watermark Claude text worldwide.2
What it means for you
If you use ChatGPT in the EU, your text will carry the mark once the rollout reaches your account. Outside the EU, nothing changes by default for now.
Fact ledger
How labels workQuestions readers ask
Will ChatGPT text be watermarked in India?
Not by default. OpenAI says the ChatGPT and Codex watermark applies to users in the EU at launch. API users anywhere can choose to turn it on.
Can anyone check if text came from ChatGPT?
Not yet. OpenAI is giving detector access first to approved researchers and expert organisations.
Does the watermark identify me?
OpenAI says no. The detector only says whether an OpenAI watermark is present.