Archive · 7 Oct 2026
OpenAI Plans Invisible Watermarks for ChatGPT Text in Europe
OpenAI plans to add a hidden signal to some text made by ChatGPT and Codex in the EU.
The signal will not look like a label that readers can see.
Instead, the AI will make tiny changes to how it chooses words.
A special detector can look for the pattern in a long enough piece of text.
OpenAI says this is meant to help follow European rules about identifying AI-made content.
Some developers around the world can choose to turn on watermarking for certain models.
At first, only approved researchers and expert organisations will be able to use the detector.
Editing, translating, or using a very short passage can make the signal harder to find.
The signal cannot tell who wrote the text, how much a person helped, or whether the text is true.
OpenAI plans to watermark eligible ChatGPT and Codex text for EU users over the coming weeks.
Its textGrain system subtly changes word-selection probabilities to create a hidden, machine-detectable pattern.
The move is intended to meet EU AI Act requirements for identifying AI-generated content.
Selected API models will offer watermarking worldwide as an opt-in; detector access will initially be limited to approved researchers and expert organisations.
OpenAI says edits and some text types can weaken detection, and the watermark cannot establish authorship, human contribution, accuracy or responsibility.
- Who
- OpenAI, with eligible ChatGPT and Codex users in the EU and developers who opt in for selected API models.
- What
- OpenAI plans to watermark eligible AI-generated text and initially restrict detector access to approved researchers and expert organisations.
- When
- The EU rollout is expected over the coming weeks; API customers can opt in, and expert organisations can apply for detector access.
- Where
- Eligible ChatGPT and Codex users in the EU; selected API models can use the feature worldwide by opt-in.
- Why
- To support EU AI Act requirements that AI-generated content be identifiable by machines.
Potential benefits
Limitations and concerns
Machine-readable identification
Potential benefits
OpenAI says the watermark can help systems identify text generated or processed by its models and support transparency requirements.
Limitations and concerns
The watermark is a limited signal; it does not identify the user, establish authorship or show how much a person contributed.
Detector access
Potential benefits
Approved researchers and expert organisations can evaluate reliability and investigate responsible uses.
Limitations and concerns
The detector will not initially be available to the general public, and edits or certain kinds of text can reduce detection reliability.
Meaning of a detection result
Potential benefits
A positive result may indicate that an OpenAI system generated or processed part of a passage.
Limitations and concerns
The signal cannot measure human judgment or creativity, establish responsibility, or indicate whether the text is accurate; a negative result does not prove a human wrote it.
- Watermark method
- textGrain subtly influences word-selection probabilities to create a statistical pattern.
- Products
- Eligible text generated by ChatGPT and Codex.
- EU rollout
- Planned over the coming weeks for eligible users; articles differ on whether rollout has already begun.
- API availability
- Selected models worldwide, with developers required to opt in.
- Detector access
- Initially limited to approved researchers and expert organisations, with access assessed case by case.
- Editing test
- Replacing about 10% of words with synonyms reduced reported detection from approximately 92% to 66%.
- Limitations
- Short passages, mathematical text, translations and editing can challenge detection; the watermark does not establish authorship, human contribution, accuracy or responsibility.
Sources
OpenAI to add invisible watermarks to ChatGPT text in EU to meet AI act rules
OpenAI plans to watermark ChatGPT and Codex text output in EU amid AI transparency push
ChatGPT Will Soon Watermark AI-Generated Text In Europe: Here Is Why
OpenAI plans to bring hidden watermark for ChatGPT and Codex