OpenAI is preparing to add invisible watermarks to eligible text generated by ChatGPT and Codex in the European Union, as part of its response to the EU AI Act’s transparency requirements. The rollout is expected to begin over the coming weeks and will apply to eligible users across ChatGPT and Codex plans in the EU.
The technology, called textGrain, works by subtly influencing the model’s choice between possible words or word pieces. This creates a statistical pattern within the text that a specialised detector can analyse. The watermark is not visible to readers and does not rely on hidden spaces, characters or unusual punctuation.
OpenAI said the move is intended to support the EU AI Act’s requirement that AI-generated content be marked in a machine-readable format so that it can be detected by other systems. The relevant transparency obligations under Article 50 began applying on August 2, 2026.
However, the watermark will not function as a foolproof AI detector. OpenAI’s testing found that shorter passages are harder to identify and that editing text can significantly weaken the watermark signal. The company therefore plans to initially restrict access to its text detector to approved researchers and expert organisations.
The watermark also cannot identify the person who generated a passage, determine how much human input was involved or establish ownership of the text. Conversely, the absence of a detected watermark does not prove that content was written entirely by a human.

AdvertisementThe Puranic