OpenAI will begin watermarking text generated by ChatGPT and Codex for users in the European Union, according to source reports from TechCrunch and The Verge. The change is tied to the EU AI Act, which pushes providers of AI-generated content to make machine-produced material easier to identify.
The system described in the saved source material uses invisible, machine-readable signals rather than visible labels. TechCrunch reported that OpenAI has been developing watermarking approaches for text, while The Verge connected the move to OpenAI’s wider compliance work around EU rules for AI systems and generated content.
The rollout also places OpenAI alongside other AI labs that are testing or deploying provenance tools. The Verge noted Google DeepMind’s SynthID as a related approach, and the source material also referenced Anthropic’s work on text watermarking. The broader goal is to help platforms, publishers, educators, and regulators distinguish AI-generated writing from human-written material when that distinction matters.
The limits are equally important. TechCrunch’s summary noted that editing generated text can make invisible watermark signals harder to detect. That means watermarking may help with large-scale detection and compliance, but it is unlikely to act as a perfect proof system for every copied, rewritten, or heavily edited passage.
For readers, the practical effect should be quiet: ChatGPT and Codex output may carry additional behind-the-scenes metadata or patterns in the EU, while the visible text experience remains largely the same. For publishers and platforms, the change signals that AI provenance is becoming part of the product layer rather than a separate research topic.
