OpenAI Introduces EU Text Watermarking
OpenAI is introducing invisible watermarking for text generated by ChatGPT and Codex within the European Union to comply with the transparency mandates of the EU AI Act, according to a corporate announcement reported by TechCrunch. The feature is rolling out over the coming weeks to eligible users across all subscription plans in the region, while global developers utilizing OpenAI APIs can enable the setting manually.
The underlying mechanism, named textGrain and detailed in a research paper co-authored with academics from the University of Pennsylvania and Yale, relies on subtly adjusting the probability distribution of word choices during generation. This leaves an imperceptible statistical pattern that specialized detectors can identify even after text is copied and pasted. OpenAI confirmed that performance metrics for the underlying models remained stable during testing.
Despite the rollout, the company acknowledged several limitations in its technical framework. Internal testing demonstrated that replacing roughly 10 percent of words with alternative synonyms reduced detection success rates from 92 percent to 66 percent. Shorter text passages, mathematical outputs, and translated content also present detection challenges. Because of these constraints, OpenAI is restricting initial access to detectors to approved researchers and expert organizations to evaluate reliability.
The move follows a similar announcement from competitor Anthropic, which implemented text watermarking globally for Claude users. OpenAI previously delayed deploying its completed text watermarking technology over competitive concerns regarding user retention. Major enterprise players including Microsoft, Google, and Meta have also committed to aligning with the regulatory framework governing artificial intelligence transparency in European markets.
Based on reporting by techcrunch.com.
