OpenAI to Add Invisible Watermarks to ChatGPT and Codex Text Under EU AI Act Rules

OpenAI has announced plans to add invisible watermarks to text generated by ChatGPT and Codex as part of its efforts to comply with the European Union’s AI Act transparency requirements. The watermarking feature is expected to roll out over the next few weeks for eligible ChatGPT and Codex users across all plans, but the initial rollout will be limited to the EU. OpenAI API developers worldwide will also be able to enable the feature for selected models.

The technology, called textGrain, adds an invisible statistical signal through the model’s choice of words, which can later be analysed by a detector. OpenAI says the system can help identify AI-generated text but acknowledges that detection has limitations. Short or highly constrained text can be difficult to detect, while editing AI-generated content may weaken the watermark. The company also stressed that a watermark does not identify the user, establish ownership or responsibility, determine accuracy, or prove that content was created entirely by AI. The move follows similar watermarking efforts by Anthropic for text and files generated by its Claude models.