ChatGPT users in the European Union are about to see a major change that most people will never actually see. OpenAI has announced that it will begin adding an invisible watermark to ChatGPT and Codex text for eligible users in the EU over the coming weeks, as the company responds to transparency requirements under the EU AI Act.
ChatGPT Watermark: What exactly is OpenAI changing?
The new system, called textGrain, does not put a visible logo, stamp or special symbol on ChatGPT answers. Instead, it subtly changes the statistical pattern of word choices while the text is being generated. A detector can then look for that pattern.
This would mean that when someone searches for ChatGPT online, ChatGPT free, ChatGPT login, ChatGPT app, or simply uses ChatGPT within the European Union, they will not be greeted with a watermark being printed on their generated response.
Why is ChatGPT getting a watermark in Europe?
The initiative is part of the European Union’s Artificial Intelligence Act, according to which all providers of generative AI have to provide an option to identify AI-generated content in a machine-readable form. OpenAI says that at launch, the rollout will apply to eligible ChatGPT and Codex users on the EU subscriptions.
For users outside Europe, there is no equivalent automatic ChatGPT text rollout announced. OpenAI says API customers worldwide can opt into text watermarking for selected models, while the setting remains off by default.
Can ChatGPT watermark detection be trusted completely?
This is where the story gets interesting. OpenAI itself acknowledges that text watermark detection has limitations. In its testing, changing 10% of words with synonyms reduced detection from about 92% to 66%; changing 25% reduced it to about 17%. Shorter and highly constrained text can also be harder to detect.
OpenAI is therefore not opening its detector to everyone immediately. Access will initially be limited to approved researchers and expert organisations. The company also stresses that detecting a watermark does not reveal the user, their prompt or their conversation—and a missing watermark does not prove that a human wrote the text.
The development is part of OpenAI’s broader push for AI content provenance, alongside Content Credentials and invisible signals already used for supported images and audio.
For more AI and technology coverage, readers can follow DemocraX Technology. For the primary announcement, see OpenAI’s explanation of its EU text-provenance approach and the OpenAI Help Center’s guide to provenance signals.
The biggest takeaway is simple: the ChatGPT watermark will be invisible to users, but potentially visible to machines.
Want to stay updated on the biggest stories shaping India and the world? Explore more on DemocraX
