OpenAI moves to watermark ChatGPT text in the EU
OpenAI says it is implementing an invisible watermark for ChatGPT-generated text in the EU, but the timing and extent of the roll-out remain unclear.
Review information is loading.
Dutch news agency ANP reported on 6 October that OpenAI would introduce the watermark over the following weeks. OpenAI’s help page describes its EU ChatGPT text watermarking in the present tense, but gives no start date for the roll-out. The company says coverage can vary by product, model, export route and when the text was created. [1] [2]
The system, called textGrain, subtly changes the pattern of word choices as ChatGPT writes. It does not add hidden characters or spaces. Detecting the pattern requires the right key and settings, and OpenAI says access to its text detector is currently limited to approved research and academic organisations. [2]
OpenAI says short passages and answers with little room for alternative wording can be difficult to detect. Substantial rewriting or translation can also weaken the signal. Finding a watermark would not establish how much of a text a person wrote. [2]
The move is intended to meet EU AI Act rules on machine-readable marking of AI-generated content. The European Commission says those rules have applied since 2 August 2026, with a grace period until 2 December only for systems placed on the market before 2 August. Separate rules can require visible labels on certain AI-generated text about matters of public interest. [2] [3]
Sign in to give this article a thumbs up or down.
About this discussion and AI questions
Private test discussion. Comments are readers’ views and are not yet automatically fact-checked. Editing is available for 60 seconds after posting.
Sign in with a confirmed reader account and choose a username to read comments and participate.
Sorting applies to top-level comments; replies remain oldest first. New comments and likes can change the order. Refresh for the current ranking.
Loading comments…