OpenAI to Add Invisible Watermark to ChatGPT and Codex Text in the EU
OpenAI said Monday it will roll out an invisible statistical watermark, called textGrain, to eligible ChatGPT and Codex output in the European Union over the coming weeks to comply with EU AI Act transparency rules, while API developers worldwide can opt in starting today.
OpenAI will begin adding an invisible watermark to text generated by ChatGPT and its coding tool Codex for users in the European Union, the company announced Monday, according to TechCrunch.
The rollout, planned over the coming weeks for eligible users on all plans in the EU, is intended to comply with transparency rules under the EU AI Act that took effect Aug. 2. Those rules require AI-generated content to be marked in a way other systems can identify, TechCrunch reported.
How the watermark works
OpenAI calls the method textGrain. Rather than inserting a visible symbol, it subtly shapes word choices so a pattern remains in the text even after copying and pasting, according to TechCrunch. The company said the mark does not identify individual users and did not measurably change model performance in its testing.
Developers using the OpenAI API anywhere in the world can turn the feature on for select models starting today, though it is off by default. OpenAI said it is not making watermarking a global default at launch.
Limits and detector access
OpenAI cautioned that editing can weaken detection. In one test cited by TechCrunch, replacing about 10 percent of words with synonyms reduced detection from roughly 92 percent to 66 percent, and short passages, math answers and translated text were harder to detect.
Access to the detector will initially be limited to approved researchers and expert organizations, TechCrunch reported. OpenAI also noted that the absence of a watermark does not by itself prove a passage was written by a person, since text could be too short, heavily edited, or produced by another system.
Reporting based on coverage by TechCrunch.
