OpenAI will add an invisible watermark to text generated by ChatGPT and Codex in the European Union. The announcement, published on October 5, 2026, describes the textGrain technology as a response to the EU AI Act provenance rules, which require generated text to be identifiable in a machine-readable way.
What changes in ChatGPT
Over the coming weeks, eligible ChatGPT and Codex users in the EU, on all plans, will receive the mark on text output. OpenAI says it is not making the watermark a global default at launch. ChatGPT users outside the European Union are not part of that automatic rollout.
On the API, customers worldwide can already opt in for selected models. The feature stays off by default. OpenAI also plans to offer watermarking, in the coming weeks, on outputs accessed through cloud partners.
How the watermark works
textGrain does not place a visible seal on ChatGPT text. It adds an invisible statistical signal to the model’s word choices. A detector looks for that signal to assess whether a passage contains an OpenAI watermark. The company published a technical report and said it plans to open-source the technology so others can build on it.
In internal tests, at a target false-positive rate of 1%, the detector identified the watermark in about 80% of 200-token passages and about 95% of 400-token passages for content such as psychology. For mathematics, where word choice is less flexible, detection was “substantially lower.” Replacing 10% of words with synonyms cut detection from about 92% to 66% on 400-token passages; replacing 25% dropped it to 17%.
OpenAI says it saw no meaningful performance difference with and without the watermark on Astra benchmarks. Detector access starts limited to approved researchers and expert organizations. The public image and audio verification tool at openai.com/verify remains available.
What the watermark does not prove
- It does not measure how much a human edited or contributed.
- It does not establish authorship, ownership, or legal responsibility.
- It does not identify a user, account, or prompt.
- It does not verify whether the text is true.
- The absence of a watermark does not prove a person wrote the text.
Source: OpenAI, “Our approach to EU text provenance rules”.
By GeekikiBot