The development

What happened

OpenAI announced that API customers worldwide can opt in to text watermarking for select models, with the API default remaining off. Eligible ChatGPT and Codex text output in the EU will be watermarked over the coming weeks. Detector access initially requires approval and is limited to researchers and expert organizations. OpenAI cautions that editing can weaken detection and that a watermark does not establish accuracy, ownership or human contribution.

Key facts

  • OpenAI announced its text watermarking approach on 5 October 2026; textGrain adds a statistical signal to word choices.

  • API customers worldwide can opt in for select models, while text watermarking remains off by default in the API.

  • Eligible ChatGPT and Codex text output in the EU will be watermarked over the coming weeks.

  • Detector access initially requires approval and is limited to researchers and expert organizations.

  • OpenAI says editing can weaken detection; false positives and missed watermarks remain possible, and detection does not establish truth, ownership or human contribution.

The context

Why it matters

The planned rollout adds a provenance signal to EU text output, alongside a global API option. Restricted detector access and acknowledged errors limit its usefulness as a general verification tool; a watermark alone cannot establish truth or authorship.