October 7, 2026
European ChatGPT users are beginning to see a change they will probably never notice directly.
OpenAI has started rolling out an invisible, machine-readable watermark for AI-generated text produced by ChatGPT and Codex in the European Union. The technology is called textGrain.
The text still looks completely normal.
There is no visible label saying “Written by AI”.
Instead, OpenAI embeds a statistical pattern into the generated text that specialized detection systems can look for later.
Why is OpenAI doing this?
The move is connected to Europe’s AI Act.
The EU’s transparency requirements for certain AI-generated content began applying in 2026, including requirements around machine-readable markings for synthetic content.
OpenAI’s watermarking system is therefore part of a broader European push to make AI-generated material easier to identify.
The technology is initially being introduced gradually.
It also affects Codex
The watermark is not limited to ordinary ChatGPT conversations.
OpenAI is also introducing the technology for Codex, its AI coding agent. That means AI-generated programming output can potentially carry the same type of provenance signal.
This is becoming increasingly relevant as AI-generated code becomes common in professional software development.
A company may eventually want to know not only what code was produced, but also where it came from and whether AI was involved.
It does not prove who wrote something
There is an important limitation.
A watermark does not prove that a particular person used ChatGPT.
Nor does it determine exactly how much of a document was generated by AI.
OpenAI also acknowledges that editing, translation and short pieces of text can weaken the ability to detect the watermark.
Therefore, the technology should not be treated as a perfect AI detector.
That matters particularly for schools and workplaces.
A detection result alone should not automatically become proof of misconduct.
Europe is becoming a testing ground
The European Union is increasingly becoming one of the most important markets for AI transparency technology.
The AI Act is being implemented progressively. Many of its major provisions became enforceable from August 2, 2026, while additional rules will take effect over the coming years.
As a result, companies such as OpenAI and Anthropic are adapting their products specifically for European requirements.
That means European users can sometimes receive AI features that are different from those available elsewhere.
What does this mean for everyday ChatGPT users?
For most people, almost nothing changes.
You can continue writing prompts and receiving answers in exactly the same way.
The difference happens behind the scenes.
If someone later uses compatible detection technology, it may be possible to identify that a piece of text contains OpenAI’s provenance signal.
That could become increasingly important as AI-generated content spreads across education, journalism, business and social media.
The larger trend is clear.
Europe is pushing AI companies toward a future where generated content is not only powerful, but also traceable.
Sources:
The Verge – OpenAI adds text watermarking to ChatGPT and Codex
European Commission – AI Act enforcement framework
Laat een reactie achter