The watermark will roll out over the coming weeks to eligible ChatGPT and Codex users in Europe. It is not a visible mark but a statistical pattern embedded in word choices, designed to let detectors identify AI-generated text. OpenAI published a technical report for textGrain alongside the announcement, co-written with university researchers.
The company was candid about the limits. Replacing about 10 percent of the words can remove the watermark, detection access is restricted to approved researchers, and the absence of a mark does not prove human authorship. The disclosure matters because it sets realistic expectations for a technology that regulators are increasingly treating as a safeguard.
OpenAI is following, not leading. Anthropic began watermarking Claude-generated text worldwide two months ago and faced user backlash. Google, Meta, Microsoft and OpenAI have all committed to the EU’s voluntary code of practice on AI-generated content, and OpenAI had built a watermark earlier but held it back, according to 2024 Wall Street Journal reporting. Separately, OpenAI began testing visual ads alongside ChatGPT image-generation results (TechCrunch, October 5), a monetization step worth watching as the company burns capital on compute.