OpenAI plans default text watermarking in the EU

OpenAI will introduce invisible watermarking for eligible ChatGPT and Codex outputs across all plans in the European Union over the coming weeks. Announced on 5 October 2026, the rollout will make the EU the company’s first region with watermarking enabled by default. API customers worldwide will be able to opt in for selected models, with the feature remaining off by default.

The technology, called textGrain, embeds a statistical signal through word choices that a specialised detector can assess. OpenAI plans an updated technical report and an open-source release, but detector access will initially be restricted to approved researchers and expert organisations. At a target false-positive rate of 1%, evaluations detected watermarks in about 80% of 200-token psychology passages and about 95% of 400-token passages. Performance was weaker for mathematics, and editing reduced detection.

The rollout follows the application of the EU AI Act’s Article 50 transparency obligations on 2 August 2026. These require machine-readable marking of synthetic outputs where technically feasible, alongside separate disclosure duties for deployers. A detected watermark does not establish human contribution, ownership, responsibility or accuracy, and an absent signal does not prove human authorship. Organisations should distinguish provider marking from deployment disclosures and document how generated text is reviewed, edited and published.