OpenAI adds invisible watermarks to ChatGPT text in the EU

OpenAI has announced it will begin embedding invisible watermarks into text generated by ChatGPT and Codex within the European Union. The move is designed to comply with the EU AI Act's transparency requirements, which went into effect on August 2 and mandate that companies label synthetic content. While the watermarking feature will roll out to all EU subscription tiers in the coming weeks, OpenAI is not making it a global default. However, developers worldwide using OpenAI's API can manually enable the feature for select models starting today.

The underlying technology, dubbed textGrain, was developed in collaboration with researchers from Yale University and the University of Pennsylvania. Instead of inserting a visible symbol, textGrain uses a secret key to subtly influence next-word predictions during text generation. This creates a hidden pattern that persists even when the text is copied and pasted. OpenAI claims the watermarking process does not compromise model performance and does not track individual user identities.

For practitioners and developers, the system is not foolproof. OpenAI's technical report reveals that simple editing can bypass the watermark; substituting just 10% of the generated words with synonyms caused the detection rate to plummet from 92% to 66%. Furthermore, the watermark is less reliable on short text blocks, mathematical solutions, and translations. Because of these limitations, OpenAI is restricting access to its detection tool to select researchers and expert groups, warning that a clean scan "does not prove human authorship."

This rollout represents a major shift for OpenAI, which previously delayed releasing its watermarking technology out of concern that users would migrate to competitors. The landscape is shifting, however, as Anthropic recently implemented global watermarking for its Claude models, despite facing pushback from users who argued the tool downplayed their creative input. Along with other tech giants like Google and Microsoft, OpenAI has pledged to align with the EU's code of practice on synthetic media.