OpenAI rolls out text watermarking for EU users and API customers

OpenAI is implementing text watermarking for users of ChatGPT and Codex within the European Union. Furthermore, API customers globally can now elect to receive watermarked text outputs for certain models starting today.
OpenAI's textGrain watermarking technique reportedly matched or surpassed other methods, such as Google DeepMind’s SynthID, when testing text. The adoption of this measure by both OpenAI and Anthropic stems from the need to satisfy the stipulations of the EU’s AI Act.
Anthropic had previously announced its watermarking efforts in August. Researchers and expert bodies can begin applying for detector access starting today.
Access to these detectors for researchers will initially be granted following a case-by-case assessment. It remains clear that text watermarks do not guarantee reliable detection, nor do they verify accuracy, determine text ownership, measure human contribution, or prove human authorship.




















































































