OpenAI is adding text watermarking in ChatGPT and Codex — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- OpenAI is rolling out "textGrain" watermarking to ChatGPT and Codex for eligible EU users across all plans over the coming weeks, but is not making it a "global default" at launch.
- API customers worldwide can opt in to watermarked text outputs for select models starting today, with cloud partners gaining access in the coming weeks.
- OpenAI says textGrain "matched or exceeded" Google DeepMind's SynthID for text — the same approach Anthropic adopted in August to meet the EU AI Act.
- OpenAI published AI benchmark scores showing similar performance between watermarked and unwatermarked text, while noting textGrain "does not guarantee reliable detection."
- OpenAI explicitly states watermarks don't verify accuracy, determine text ownership, measure human contribution, or prove human authorship.
- Approved researchers and expert organizations can apply for detector access starting today, granted case-by-case under the EU Code of Practice, though the tool will not be publicly available at launch.
Why it matters: OpenAI is threading EU regulatory compliance without globally defaulting to watermarking, leaving non-EU markets untouched while API customers self-select transparency. The explicit disclaimers — that watermarks don't verify accuracy, authorship, or human contribution — pre-empt misuse of the signal as a content-authenticity certificate, a distinction downstream platforms and regulators will need to internalize.
Ask SkimNews
