OpenAI will add an invisible digital watermark to every text and piece of code its models generate within the European Union. The new policy aims to meet EU guidelines on transparency obligations for AI providers and deployers. Kursiv LifeStyle explains how it will work.
To comply with the new requirements, OpenAI will use its textGrain technology, which alters text during generation without being noticeable to the user. While such text looks completely normal, specialized software can detect — based on specific characteristics — that it was written by AI. The watermark itself is invisible to the naked eye.
Once OpenAI activates this new labeling, it will automatically apply an invisible watermark to every text and piece of code ChatGPT generates within the EU. Outside Europe, certain OpenAI models are also expected to offer this option, but users will decide whether to enable or disable it.
The watermark will not reveal a specific user’s identity, nor will it expose their prompts or chat history. Initially, only approved researchers and specialized organizations will be granted rights to verify texts. The company, however, plans to make the technology available to the general public in the future.
Under EU regulations, AI providers must ensure users are explicitly informed whenever they interact directly with an AI system and add machine-readable marks to enable the detection of content generated or manipulated by AI. The deadline for compliance with the new rules is Dec. 2.
