Morph1k visual note · 01OpenAI starts EU text watermarking with textGrain
OpenAI published its EU text-provenance plan: opt-in textGrain watermarking for select API models globally starting October 5, invisible marks on eligible ChatGPT and Codex text in the EU over coming weeks, and detector access limited at first to approved researchers.
Primary reading
OpenAI: Our approach to EU text provenance rulesWhat happened
On October 5 OpenAI described how it will meet EU AI Act machine-readable text requirements with textGrain, a statistical watermark woven into word choice rather than hidden characters. API customers worldwide can opt in for select models; watermarking stays off by default in the API. Over the coming weeks, eligible ChatGPT and Codex text output in the European Union will carry an invisible watermark across all plans — not as a global default. Detector applications open for approved researchers and expert organizations; image and audio verification tools remain publicly available. OpenAI’s evaluations show detection improving with length (about 80% of 200-token vs ~95% of 400-token psychology passages at a 1% false-positive target) and weakening under edits (10% synonym replacement cut detection from ~92% to 66% on 400-token samples). It plans to open-source textGrain and reports no meaningful Astra benchmark drop with watermarking on.
Why it matters
Authorship fights used to end in a shrug. A major lab is now putting a machine-readable mark into everyday EU chat and coding text — while saying aloud what the mark cannot prove about ownership, accuracy, or how much a human rewrote.
For you
If you ship in the EU or sell to EU users, expect ChatGPT and Codex drafts to carry an invisible OpenAI mark soon — and treat a missing detection as inconclusive. API builders who need provenance can opt in now; everyone else should keep human disclosure and version history, because light edits erase the signal fast.
