OpenAI sets out its text watermarking approach for the EU AI Act
OpenAI set out its text watermarking approach in response to the EU AI Act — the statistical watermark textGrain is opt-in for select models in the API (off by default), and an invisible watermark will be added to ChatGPT and Codex text output in the EU within weeks
Why it matters
The approach targets the EU AI Act requirement to mark generated text in a machine-readable way, and starts in the EU only rather than as a global default. The detector opens case by case to approved researchers and expert organizations, and OpenAI reported no meaningful performance difference with and without the watermark on Astra benchmarks
Source — Primary sources first; secondary reports are labelled “Press”
Officialopenai.comhttps://openai.com/index/eu-text-provenance/Added in Week 42, 2026 · See that week