Policy
OpenAI will watermark ChatGPT and Codex text for EU users with 'textGrain'; API developers can opt in worldwide
OpenAI said it will roll out textGrain, an invisible statistical watermark that shapes token choices, to ChatGPT and Codex output for EU users on every plan over the coming weeks, to meet the EU AI Act's content-marking rules. API customers anywhere can turn it on per project, but it stays off by default. OpenAI's technical report, written with researchers at Penn and Yale, puts detection near 92-95% on passages of about 400 tokens at a 1% false-positive rate, dropping to 66% when 10% of words are swapped for synonyms; detector access starts with approved researchers. Builders shipping to EU users should expect watermarked model output soon and should not treat a missing watermark as proof of human authorship.
