OpenAI will watermark ChatGPT outputs by default—but only in the EU
Like other solutions, it is not especially reliable, and it’s easy to circumvent.
TL;DR
- OpenAI will start automatically watermarking text generated by ChatGPT in the European Union.
- This move is to comply with the EU AI Act, which requires AI-produced content to be marked for detection.
- The watermarking method, called textGrain, is proprietary but works by embedding undetectable patterns in word choices.
- OpenAI acknowledges that the watermarking is not completely effective, with detection rates dropping significantly if the generated text is modified.
- The feature will be a default in the EU but optional in other regions and available in the API.
- Competitor Anthropic also introduced watermarking globally for its models' outputs.