OpenAI said it will introduce invisible text watermarking for eligible ChatGPT and Codex users in the European Union (EU) to comply with the EU AI Act even as the company claimed that the watermarking has limitations.
In a statement on Monday, OpenAI announced the release of textGrain, its text watermarking, to identify whether a text sample was likely generated by ChatGPT. The company said it is responding to the requirements under the EU AI Act to make AI-generated text identifiable in a machine readable way.
“The EU AI Act requires generative AI providers to make generated text identifiable in a machine-readable way. Text watermarking and detection remain early technologies with significant limitations, and views about their benefits and responsible uses are still developing,” OpenAI said.
The company said the watermarking will initially be rolled out only in the EU, while API customers globally will be able to opt in to text watermarking for select models from today. The feature will remain off by default for API customers.
OpenAI said watermarking technique is still in an evolving phase with key limitations.
” Watermarks have limits. They’re often undetectable, especially in short passages. Rewriting or translating text can completely remove the watermark. That’s why, for now, only approved researchers will have access to our detector, so they can help us evaluate and improve the technology. We’re treating text watermarking as an ongoing area of research. We’ll keep testing and improving, with feedback from users, developers, policymakers, and researchers,” said OpenAI in a tweet on X.
Watermarks have limits. They’re often undetectable, especially in short passages. Rewriting or translating text can completely remove the watermark.
That’s why, for now, only approved researchers will have access to our detector, so they can help us evaluate and improve the technology.
We’re treating text watermarking as an ongoing area of research. We’ll keep testing and improving, with feedback from users, developers, policymakers, and researchers.
https://t.co/2QUil8oWdB
— OpenAI (@OpenAI) October 5, 2026
The company is also opening applications for access to its text watermark detector, but access will initially be limited to approved researchers and expert organizations to help evaluate and improve the technology.
Announcing our work textGrain, @OpenAI’s text watermark in response to the EU AI Act. OpenAI is rolling out textGrain in the EU and opening detector access to researchers.
We’ll open-source it so the community can build on top. https://t.co/BoSnzDZ738
— Weijie Su (@weijie444) October 5, 2026
How OpenAI Text Watermarking ‘textGrain’ Works?
The company’s text watermarking technology, called textGrain, adds an invisible statistical signal to a model’s word choices. OpenAI’ detector then checks whether a passage contains the signal.
According to OpenAI’s evaluations, textGrain matched or exceeded the performance of other approaches tested by the company, including Google’s SynthID for text. However OpenAI said text watermarking and detection are still early technologies and have significant limitations.
“At a target false positive rate of 1%, our detector identified watermarks in about 80% of 200-token passages, compared with about 95% of 400-token passages, for content such as psychology. Detection rates were substantially lower for content such as mathematics, where there is less flexibility in word choice,” the company mentioned.
OpenAI also found editing can significantly weaken the watermark. In a test involving 400-token passages, replacing 10 percent of words with synonyms reduced detection from about 92 percent to 66 percent, while replacing 25 percent of words reduced detection to 17 percent.
The company said these limitations are among the reasons why it is initially restricting access to the detector.
“These limitations contribute to our decision to provide initial detector access only to approved researchers and expert organizations, who can help us evaluate reliability and responsible uses,” OpenAI said.
The OpenAI text watermarking provides only a limited signal about the role its systems played in producing or processing a passage. A watermark does not measure the extent of human contribution, establish ownership or responsibility, identify the user or verify whether the text is accurate.
Similarly the absence of a detected watermark does not prove that a text was written by a human. OpenAI said detection can fail when text is too short, heavily edited or translated. Text may also have been generated using an unsupported model, before watermarking was introduced or by another company’s AI system.
The detector will report whether it detects an OpenAI watermark but will not identify the user or reveal prompts or conversations.
OpenAI highlighted that it plans to make its text watermarking technology available as open source so that others can build on it. “We’ll continue improving detection, studying how watermarks withstand editing and translation, and exploring ways to distinguish AI assistance from AI authorship more meaningfully,” the company said.
Also Read: Claude Will Watermark Its Text Globally Under EU Rules; Key Limitation Persist
