AI NexusAI Nexus
Industry

OpenAI starts opt-in text watermarking and plans an EU rollout for ChatGPT and Codex

Event time 1 independent sourceEditorial score 83/100Updated here

The short version

OpenAI made text watermarking opt-in for selected API models and said eligible ChatGPT and Codex text in the EU will receive an invisible watermark in coming weeks.

What changed

The API setting is off by default, while detector access starts with approved researchers and expert organizations rather than the public.

What it means for you

Developers and EU users need to understand the regional rollout, opt-in API behavior, and the limits of a detection result before treating it as proof.

OpenAI published its approach to EU text provenance rules on October 5. The company says API customers worldwide can now choose text watermarking for selected models, but the API option remains off by default.

For eligible ChatGPT and Codex text output in the European Union, OpenAI plans to add an invisible watermark over the coming weeks. It describes the watermark as a statistical signal in word choices, called textGrain. The company also says a positive result cannot identify a user, prove ownership, establish factual accuracy, or measure a person's contribution to the text.

OpenAI is not opening the text detector to everyone at launch. Applications are available to approved researchers and expert organizations, while the company evaluates reliability and responsible use. Its own evaluation notes that shorter or constrained passages are harder to detect and that editing can weaken the signal.

For API teams, the immediate action is to decide whether opt-in watermarking belongs in a product's transparency workflow and to test its effect on the intended output. For EU ChatGPT and Codex users, this is a regional rollout rather than a global default. Review OpenAI's announcement before making policy or compliance claims about detector results.

Codex CLI

Fact check

  • VerifiedOpenAI announced on October 5, 2026 that API customers can opt in to text watermarking for select models, with the setting off by default.Evidence
  • VerifiedOpenAI said it will add an invisible watermark to eligible ChatGPT and Codex text output in the European Union in coming weeks, and initially limit detector access to approved researchers and expert organizations.Evidence

Coverage timeline

  1. Primary sourceOpenAI
    Our approach to EU text provenance rules