OpenAI Rolls Out textGrain Text Watermarking in ChatGPT and Codex for EU Users
OpenAI is rolling out an invisible, machine-readable text watermark called textGrain to ChatGPT and Codex, but initially only for users in the European Union. The company says the move helps it meet the requirements of the EU's AI Act, following similar steps by Anthropic and Google DeepMind, whose SynthID for text OpenAI claims textGrain matched or exceeded in testing. Benchmark scores show watermarked and unwatermarked text performing similarly.
OpenAI cautions that textGrain "does not guarantee reliable detection" and does not verify accuracy, determine ownership of the text, measure human contribution, or prove human authorship. The watermark is designed to signal that text is AI-generated, but its limitations mean it cannot serve as definitive proof of a document's origin.
The rollout will reach eligible ChatGPT and Codex users across all plans in the EU over the coming weeks, but OpenAI says watermarking will not become a global default at launch. The company is taking a regional approach to learn from real-world use and feedback, while API customers around the world can opt in to watermarked outputs for select models starting today.