October 6, 2026
OpenAI is beginning to introduce an invisible watermarking system for text generated by ChatGPT and Codex in the European Union.
The technology, called textGrain, creates a machine-readable marker within AI-generated text. OpenAI says the system is being introduced as Europe implements stricter transparency requirements for AI-generated content.
The move represents a significant change for European users.
AI-generated text has traditionally been difficult to distinguish from human writing. OpenAI’s new system attempts to create a technical signal that can be used by detection systems without visibly changing the text.
Why Europe is getting the technology first
The European Union’s AI Act entered a new phase in August 2026.
The latest transparency requirements cover certain AI-generated and AI-manipulated content. The European Commission says machine-readable markings are now required for certain categories of AI-generated material.
OpenAI’s European watermarking initiative is therefore closely connected to the EU’s regulatory approach.
Rather than asking users to disclose manually that they used AI, the technology attempts to build provenance information directly into the generated output.
The watermark is invisible
Users should not expect to see a visible logo or notice inside their ChatGPT responses.
The marker is designed to be machine-readable.
OpenAI says it has developed the system so that the watermark does not materially affect model performance. The company is also providing access to detection technology to selected researchers and organisations working on provenance research.
That could eventually make it easier for platforms, researchers and institutions to determine whether a piece of text originated from an AI system.
But watermarking is not the same as proving authorship
There is an important limitation.
An AI watermark does not necessarily prove that a person did not subsequently edit the text.
OpenAI itself acknowledges that text provenance technology remains an evolving field. Significant rewriting, translation or other modifications can weaken detection.
Therefore, the technology should not be interpreted as a perfect AI detector.
Instead, it provides another signal that can potentially help identify the origin of content.
Europe is becoming the testing ground
The development illustrates a broader European trend.
The EU is attempting to establish technical standards around AI transparency at the same time that companies are rapidly expanding generative AI.
That creates a new relationship between regulation and product development.
Companies such as OpenAI must increasingly design AI systems not only around performance, but also around requirements for transparency, provenance and accountability.
For European users, that could mean more visible differences between AI products inside and outside the EU.
Sources:
The Verge – OpenAI’s AI text watermarking in the EU
European Commission – Artificial Intelligence and transparency rules
Laat een reactie achter