OpenAI Begins Phased Text Watermarking Under EU AI Act Rules

OpenAI announced on October 5, 2026 that it has begun a phased rollout of text watermarking technology for its ChatGPT and Codex models in the European Union.

The move comes as OpenAI seeks to comply with the EU AI Act's transparency requirements, which mandate that providers of generative AI systems make AI-generated text identifiable in a machine-readable form.

Starting October 5, API customers worldwide can opt in to text watermarking for select models, though the feature remains off by default.

Over the coming weeks, OpenAI will add invisible watermarks to eligible ChatGPT and Codex text output in the European Union, with the watermark detector initially available to approved researchers and expert organizations.

The company's text watermarking system, called textGrain, places an invisible statistical signal within word choices to detect AI-generated text without significantly impacting meaning or readability.

This development positions OpenAI alongside other major AI providers like Anthropic, which have also implemented marking systems to comply with EU AI Act requirements.