OpenAI is adding hidden digital markers to text created by ChatGPT and its coding tool Codex, but only for people using the services in the European Union. The change is designed to help the company follow the EU's new AI Act, a strict set of rules for artificial intelligence systems operating in Europe. The law demands that companies make it clear when content is produced by AI, and watermarking is one way to meet that requirement without changing what users see on their screens.
The watermarking system, which OpenAI calls textGrain, embeds invisible signals inside the text that machines can read but humans cannot see. According to the company, this method works as well as or better than similar tools from other tech giants, including Google DeepMind's SynthID for text. Another major AI lab, Anthropic, announced its own watermarking plan earlier this year for the same legal reason. These systems all aim to solve the same problem: giving regulators, platforms, and readers a way to identify machine-written material at scale.
OpenAI says the watermarks will not change how the AI performs on standard tests, meaning the quality of answers should stay roughly the same. However, the company admits that if a user heavily edits the text, the hidden marks may become harder to spot. To help with this, OpenAI plans to give researchers access to detection tools first, allowing outside experts to study how well the system holds up in real-world conditions.
This move matters because the EU AI Act requires clear transparency around AI-generated content. As AI writing becomes harder to tell apart from human writing, regulators want a reliable way to check where text came from. The rollout in Europe could serve as a test case for similar rules in other parts of the world. If the watermarking system works well and does not hurt the user experience, OpenAI may expand it beyond the EU. That would set a new global standard for labeling AI content and could change how all major AI companies approach transparency in the future.