OpenAI will watermark ChatGPT and Codex text for EU users under the AI Act

OpenAI will start adding an invisible watermark to text generated by ChatGPT and Codex in the European Union to comply with the EU AI Act, the company said on 5 October 2026, according to TechCrunch. The act's transparency rules, which took effect on 2 August, require AI companies to mark AI-generated content in a way other systems can identify, TechCrunch reported.

How the watermark works

• The technique, called textGrain, subtly shapes the model's word choices, leaving a pattern readers cannot see but a detector can pick up, OpenAI said, according to TechCrunch.
• The watermark travels with the text when copied and pasted and does not identify the user, TechCrunch reported.
• Editing undermines detection: replacing 10% of words with synonyms dropped detection from about 92% to 66% in one test, according to TechCrunch.
• OpenAI said it saw no meaningful change in its models' performance with the watermark switched on, TechCrunch reported.
• Detector access is, for now, limited to "approved researchers and expert organizations, who can help us evaluate reliability and responsible uses," OpenAI said, as quoted by TechCrunch.
• A watermark "can indicate that an OpenAI system generated or processed part of a passage, but not how much human judgment, editing, or creativity went into it," and "does not prove human authorship," OpenAI said, according to TechCrunch.

Rollout

• The watermark will roll out over "the coming weeks" to eligible ChatGPT and Codex users on all plans, but only in the EU, TechCrunch reported.
• API developers anywhere could turn it on for select models from the day of the announcement; it stays off by default outside the EU, according to TechCrunch.
• Anthropic, Google, Meta, Microsoft and OpenAI have committed to following the EU's code of practice on AI-generated content, TechCrunch reported.

Why it matters

This is one of the first concrete technical steps a leading US AI developer has disclosed since the EU AI Act's transparency rules took effect in August, and it shows the gap between the law's intent and what watermarking can currently guarantee: modest paraphrasing cuts detection sharply, by OpenAI's own account. For businesses and public bodies outside the EU, including in the Caspian region, that use US-made AI tools, the rollout signals that EU content-provenance duties may increasingly follow the product, shaping default settings for anyone serving EU markets even while the feature stays optional elsewhere.