OpenAI is set to automatically watermark the text generated by ChatGPT in the European Union, in line with the EU AI Act. The move is part of a broader strategy that will offer the watermarking feature in other regions, but will keep it off by default.

The EU AI Act, which came into effect in August, mandates that AI-generated content must be marked in a way that can be detected by another tool. While a few standards, like SynthID and the C2PA project, already exist, they are relatively easy to bypass for those with basic know-how.

OpenAI’s watermark, called textGrain, works by inserting patterns into word choices that are not noticeable to a human reader but can be identified by a specialized detector. The company plans to distribute the detector to a limited number of researchers and organizations and will offer a request-for-approval process for others to gain access over time.

The effectiveness of such watermarking remains a topic of debate, as no method is yet foolproof. However, this step by OpenAI in the EU might set a precedent for similar measures in other regions, highlighting the increasing need for transparency in AI-generated content.