OpenAI to add invisible text watermarks to ChatGPT and Codex for EU users

Mixed · 2OpenAI said on Monday 5 October 2026 that it is adding an invisible, machine-readable watermark to text produced by ChatGPT and Codex, initially only for users in the European Union. 1 2 Mixed · 2The rollout will reach eligible users on all plans over the coming weeks, and OpenAI says it is not making watermarking a global default at launch, describing the regional approach as a way to learn from real-world use and feedback. 1 2 Mixed · 2API customers anywhere in the world can opt in to watermarked output for select models starting immediately; it is off by default, and OpenAI is working with cloud partners to offer it through their services in the coming weeks. 1 2

Mixed · 2TechCrunch reports that the EU AI Act's transparency rules, which took effect on 2 August, require AI companies to mark AI-generated content in a way other systems can identify; The Verge likewise says OpenAI is acting to meet the Act's requirements. 2 1 Centre · 1The method, called textGrain, subtly shapes the model's word choices using a secret key, leaving a pattern readers cannot see but a detector can find. Because it lives in the wording, it travels with copied and pasted text. OpenAI published a technical report co-written with researchers from the University of Pennsylvania and Yale. 2 Mixed · 2OpenAI says textGrain matched or exceeded other approaches such as Google DeepMind's SynthID for text, and that benchmark scores for watermarked and unwatermarked text were similar. TechCrunch says OpenAI saw no meaningful change in model performance, and that the watermark does not identify the user. 1 2

Mixed · 2OpenAI acknowledges limits: textGrain "does not guarantee reliable detection", and editing can weaken it. In one test, replacing 10% of words with synonyms cut detection from about 92% to 66%, and short passages, math answers and translated text are harder to detect. 1 2 Mixed · 2OpenAI also says a watermark does not verify accuracy, establish ownership, measure human contribution or prove human authorship, and that a missing watermark does not prove a human wrote the text. 1 2 Left · 1Detector access is not public at launch because of the risk of missed watermarks and false positives. Approved researchers and expert organisations can apply from now, with access granted case by case under the EU Code of Practice, and the tool reports only whether an OpenAI watermark is detected, without identifying the user or revealing prompts. 1

Mixed · 2Anthropic announced watermarking for Claude text two months earlier, applying it worldwide, also built on SynthID according to The Verge. TechCrunch says that move drew backlash from some Claude users, who argued they had supplied the instructions and decisions and Claude was just the tool. 2 1 Centre · 1TechCrunch adds that OpenAI had built a text watermark before but held back, partly over fears users would move to rivals that did not watermark, as the Wall Street Journal reported in 2024. Anthropic, Google, Meta, Microsoft and OpenAI are among companies that have committed to the EU's code of practice on AI-generated content. 2