OpenAI outlines its approach to EU text provenance rules

OpenAI has published an official note detailing how it is approaching text watermarking under EU text provenance rules. The post lays out the company's overall thinking and how it intends to put the mechanism into practice.

According to the note, the approach begins by defining where watermarks apply—that is, which model-generated text carries an identifiable mark. For anyone tracking AI content transparency, that scope determines how much surface area provenance can actually cover.

The note also explains how detection works. The watermark does not change what a reader sees; the key sits on the detection side, where a dedicated tool is what determines whether a piece of text carries the marking.

One detail worth noting is the order of access. OpenAI said the detection capability opens first to researchers rather than to everyone at once. In the early phase, that means the research community, not ordinary users or third-party products, will be doing most of the watermark verification.

The significance is that EU text provenance requirements are turning traceability into a capability that has to be prepared in advance. By publishing its implementation path, OpenAI signals both a compliance posture and a foundation for a wider rollout later.

What to watch next is whether a detection entry point for developers and ordinary users follows the researcher phase, and whether other major model providers put forward comparable watermarking and detection plans.