OpenAI's approach to text watermarks under EU rules

OpenAI has published an explanation of how it approaches marking AI-generated text under the European Union's rules on content provenance. The material covers where watermarks apply, how detection works, and why access to verification will start with researchers rather than the general public.

A text watermark is a hidden signal that lets someone later work out that a passage was produced by a model. It is one way to tell machine-written text from human writing, and a step toward more transparency. Rolling access out in stages suggests the mechanism will stay in a narrow circle at first, with the details likely to be refined as it goes.

If you work with partners in the EU or prepare text for a European audience, draw up a simple map: which emails, product descriptions, reports and customer replies are written by AI, and who reviews them. Put a rule in your internal procedures that staff flag machine-written text, and keep an eye on EU requirements, since over time they may reach Kazakhstani companies too.