NeuralStackOpenAI will start adding an invisible watermark to text generated by ChatGPT and Codex for users in the European Union, the company said in a blog post. The feature is intended to comply with transparency requirements under the EU AI Act.
The watermark is not a visible symbol or label. Instead, OpenAI’s method subtly shapes the model’s word choices, creating a pattern that readers cannot see but detection systems can identify.
OpenAI said the watermark will roll out over the coming weeks to eligible ChatGPT and Codex users on all plans in the EU. Developers using OpenAI’s API anywhere in the world can also activate the feature for select models, although it will be turned off by default.
The company is not making text watermarking a global default at launch. OpenAI has published a technical report describing its approach, called textGrain, and said the watermark does not identify the person who generated the text.
Because the signal is embedded in the words themselves, it can remain with text when users copy and paste it. OpenAI also said it observed no meaningful change in model performance when the watermark was enabled.
Watermarking will not provide a definitive test of authorship. OpenAI’s tests found that replacing 10% of words with synonyms reduced detection from about 92% to 66%, and the company cautioned that the absence of a watermark does not prove that a person wrote the passage.
Why this matters
The rollout places OpenAI’s text-generation products among the systems adapting to the EU’s transparency rules, which require AI-generated content to be marked in a way that other systems can identify. It could help platforms, educators, and other organizations flag content produced or processed by OpenAI systems, while leaving unresolved how reliably such signals survive editing.
OpenAI said a watermark can indicate that one of its systems generated or processed part of a passage, but cannot show how much human judgment, editing, or creativity went into it. Anthropic, Google, Meta, Microsoft, and OpenAI are among the companies that have committed to following the EU’s code of practice on AI-generated content.
The initial EU-only rollout will give OpenAI room to observe how the system works under regulatory requirements before deciding whether to expand it more broadly. For now, the company’s own testing suggests that detection can weaken when text is modified, and that a missing signal cannot establish human authorship.