OpenAI will start watermarking ChatGPT text in the EU
OpenAI begins adding an invisible watermark to ChatGPT and Codex text in the EU to comply with transparency rules under the EU AI Act.
Stock photo for illustration only, not from the actual event
- OpenAI adds watermarks to ChatGPT and Codex text in the EU
- Compliance measure for the transparency rules of the EU AI Act
- Introduces textGrain technology co-authored with university researchers
- Initial detector access restricted to approved experts and researchers
OpenAI announced in a blog post on Monday, October 5, 2026, that it will begin embedding an invisible watermark into text generated by ChatGPT and Codex within the European Union. This rollout is designed to comply with the transparency mandates outlined in the EU AI Act, which officially took effect on August 2, requiring artificial intelligence companies to properly mark AI-generated outputs so external systems can easily identify them.
Stock photo for illustration only, not from the actual event
Alongside this rollout, OpenAI published a comprehensive technical report detailing its newly developed method known as textGrain. Co-authored alongside academic researchers from the University of Pennsylvania and Yale, the report outlines how a secret key is utilized to sort next-word predictions during sentence completion. By combining hundreds of these subtle adjustments, a specialized detector can reliably spot AI-generated material using only the text string and the corresponding key.
OpenAI's implementation in Europe marks a critical regulatory milestone after years of hesitation. Back in 2024, The Wall Street Journal reported that the company had previously built a text watermark but delayed its launch out of fear that users might migrate to competing platforms lacking watermarking restrictions. The enforceable mandates of the EU AI Act have now forced major technology firms, including Anthropic, Google, Meta, Microsoft, and OpenAI, to align with the region's unified code of practice for AI content.
Despite the technical achievement, the watermark system remains susceptible to various editing methods. Internal testing conducted by OpenAI revealed significant vulnerabilities regarding text alterations.
- Replacing just 10% of the words with synonyms drops detection rates from roughly 92% down to 66%
- Short textual passages
- Mathematical answers
- Translated text content
Because of these inherent limitations
Stock photo for illustration only, not from the actual event
, the company stated that initial access to the detection tool will be strictly limited to approved researchers and expert organizations to help evaluate reliability and ensure responsible deployment.
"[Watermarks] can indicate that an OpenAI system generated or processed part of a passage, but not how much human judgment, editing, or creativity went into it."
OpenAI
Source: TechCrunch
Found something wrong in this article? Report an issue with this article
Comments
Leave a Comment