OpenAI Adds Text Watermarks in EU to Meet AI Act

According to TechCrunch, OpenAI announced it is adding an invisible watermark to text generated by ChatGPT and Codex within the European Union. This rollout, slated to happen over coming weeks across all plans, satisfies transparency rules from the EU AI Act that took effect on August 2. The feature relies on a technical method called textGrain—co-written with researchers from Yale and the University of Pennsylvania—which subtly shapes model word choices using a secret key to form a pattern detectable by specialized software. Global developers using OpenAI's API can also enable this feature on select models, though it remains off by default worldwide.

This development marks a significant compliance step for major artificial intelligence developers operating in Europe. While the watermark travels with copied and pasted text and does not reveal user identities, testing shows its effectiveness diminishes through editing, synonym replacement, short passages, or translations. OpenAI's cautious initial deployment reflects ongoing industry challenges in balancing transparency regulations with user privacy, software performance, and competitive market pressures.

Anthropic previously announced plans to watermark text generated by Claude worldwide, which drew backlash from some users who argued they supplied the instructions while the AI was merely a tool.

As regulatory frameworks evolve across Europe, artificial intelligence providers continue adjusting deployment strategies to meet legal mandates while evaluating technical reliability and user feedback.