Newsbite analysis · AI, Regulation
TechCrunch
OpenAI will start watermarking ChatGPT's text in the EU
OpenAI announced it will start watermarking the text generated by ChatGPT and Codex for users in the European Union as part of its compliance with the EU AI Act. This invisible watermark will embed subtle patterns in the text that can be detected by authorized systems, maintaining user anonymity and not impacting model performance. While the watermark is exclusive to the EU for now, developers globally can activate it via the API for certain models. OpenAI emphasizes that editing the text could remove the watermark's effectiveness, cautioning that the absence of a watermark does not confirm human authorship.
MonitorConfidence: HighShort term horizonStory: First report
EvidenceOfficial announcementCompany supplied dataCompany only
Published 5 Oct 2026, 20:40 UTC3 min read
Lead image — TechCrunch
Brief
NewsBite reading:OpenAI introduces watermark for ChatGPT's text in EU compliance
TL;DR
- Watermarking will begin for ChatGPT and Codex users in the EU.
- The initiative is in response to the EU AI Act transparency rules.
- Watermarking subtly adjusts word choices to identify AI-generated content.
- Developers can enable the watermark feature through the API.
- Editing text can reduce the effectiveness of the watermark significantly.
- OpenAI aims for responsible use of the detection method among approved researchers.
Interesting facts
- The watermark is not a visible symbol but influences the wording patterns.
- Competitors like Anthropic are also implementing watermarking globally.
- Watermarked text can still be misattributed due to editing.