TLDR
- OpenAI will add an invisible watermark called textGrain to ChatGPT and Codex text in the EU.
- The watermark responds to transparency rules under the EU AI Act.
- Tests show swapping 25% of words for synonyms drops detection to just 17%.
- Only approved researchers and expert organizations can access the detector for now.
- The feature is EU-only by default, but API customers worldwide can opt in.
OpenAI announced on October 5 that it will add an invisible watermark to ChatGPT and Codex outputs across the European Union. The company says the move answers transparency requirements under the EU AI Act.
We’re expanding our approach to content provenance to include text in response to EU regulatory requirements, while recognizing the significant limitations of current text watermarking technology.
Our tools already help verify whether an image or audio file was created with our…
— OpenAI (@OpenAI) October 5, 2026
The watermark system is called textGrain. It works by nudging the model’s word choices to leave behind a hidden statistical pattern.
A separate detector tool then scans text for that pattern. OpenAI says the method adds no hidden characters, invisible spaces, or odd punctuation.
This means a common workaround, scrubbing hidden characters from text, will not remove the watermark. The signal is built into the wording itself.
How Well Does the Watermark Hold Up?
Detection accuracy depends heavily on text length. OpenAI measures text in tokens, which are word fragments that AI models process. One token is roughly three-quarters of an English word.
In company tests, the detector caught about 66% of unedited answers around 150 words long. At roughly 300 words, that rate climbed to about 92%.
Editing the text weakens the watermark fast. For 300-word passages, swapping one word in 10 for a synonym cut detection from 92% down to 66%.
Replacing one word in four pushed detection down to just 17%. OpenAI’s own data shows the system struggles once text gets edited, shortened, or rewritten.
The company also said the watermark had little effect on benchmark scores for its latest model, GPT-6 Astra.
Why Isn’t the Detector Public?
OpenAI is not releasing the detector to the general public right away. Instead, access will go to approved researchers and expert organizations.
The company says a positive match from the detector does not reveal who the user was, what they typed, or what the conversation contained.
A negative result does not prove much either. Short, edited, or translated text can slip past the detector without being flagged.
Output from other AI tools will also not trigger the system, since it only checks for OpenAI’s own pattern. OpenAI stated plainly that the absence of a detected watermark does not prove a human wrote the text.
Images and audio are handled differently than text. Anyone can already upload those files to OpenAI’s public verification tool to check for its SynthID watermark.
Outside the European Union, the ChatGPT watermark will stay off by default for regular users. API customers around the world, however, can opt in for select models through their project or organization settings.
The rollout comes after the EU began enforcing AI Act transparency rules back in August. OpenAI framed the watermark as part of its ongoing response to that enforcement.
The company also pointed to a wider data point in its announcement. Criminal use of AI tools climbed 40% year-on-year, according to research firm TRM Labs.
OpenAI did not say when or if the detector tool would open to the general public. For now, the EU rollout and limited researcher access remain the current state of the project.
Stop guessing and start investing with confidence. KnockoutStocks gives you the AI insights, market intelligence, and stock research you need to spot opportunities, cut through the noise, and make smarter investment decisions — all in one powerful platform.
Sign up today and get 50% OFF full access to our premium stock picks.
Simply use coupon code SPECIAL50 at checkout to claim your exclusive discount.
