OpenAI will embed invisible, machine-readable watermarks in ChatGPT and Codex outputs for EU users to comply with the AI Act's transparency rules.
OpenAI confirmed it will roll out its 'textGrain' watermarking system to ChatGPT and Codex for users in the European Union, marking AI-generated text with an invisible signal detectable by machines. The company says the approach matched or exceeded other watermarking methods it tested, but acknowledged heavy editing can degrade detectability.
The move is OpenAI's first concrete compliance step tied to the EU AI Act's content-labeling requirements, and it's starting regional rather than global, leaving US and other markets unmarked for now.
This is a preview of how US AI labs will fragment product behavior by jurisdiction as global AI regulation diverges. Enterprises building on OpenAI's stack in Europe need to plan for provenance requirements now, and expect similar mandates to spread to other markets with content-authenticity laws.
The daily signal, curated. Get it in your inbox.
Subscribe on LinkedIn →