OpenAI to Add Invisible Text Watermarks for EU ChatGPT Users to Meet AI Act Rules
OpenAI will roll out invisible watermarks to EU ChatGPT and Codex users over the coming weeks, covering all plans under rules effective August 2. Its textGrain detector reached roughly 92% accuracy, falling to 66% when 10% of words were replaced with synonyms.
OpenAI announced in October 2026 that it will add invisible watermarks to text generated by ChatGPT and Codex for users in the European Union, citing compliance with the EU AI Act as the reason for the move.
The EU AI Act’s transparency rules, which took effect on August 2, require AI companies to mark AI-generated content in a way that other systems can identify. OpenAI said the watermark will roll out over the coming weeks to eligible ChatGPT and Codex users on all plans within the EU. Developers using OpenAI’s API anywhere in the world can enable it for select models immediately, though it is off by default. OpenAI is not making text watermarking a global default at launch.
The watermark works by subtly shaping the model’s word choices, leaving a pattern that readers cannot see but a detector can identify. Because it is embedded in the words themselves, it travels with the text when copied and pasted. OpenAI said the watermark does not identify the user and that it observed no meaningful change in model performance with the feature enabled.
OpenAI published a technical report for its method, called textGrain, co-written with researchers from the University of Pennsylvania and Yale. Initial detector access will be limited to approved researchers and expert organizations while reliability is evaluated.
The watermark has limitations. In OpenAI’s own tests, replacing 10% of words with synonyms dropped detection accuracy from roughly 92% to 66%. Short passages, math answers, and translated text are also harder to detect. The company cautioned that a missing watermark “does not prove human authorship,” as text may be too short, too heavily edited, or generated by another company’s AI.
The announcement follows a similar move by Anthropic in August 2026, when it said it would watermark text generated by Claude globally. OpenAI had previously built a text watermark but held off releasing it, partly over concerns that users would switch to rivals that did not watermark, according to a 2024 Wall Street Journal report.