OpenAI to Watermark ChatGPT Text in the EU
In brief
OpenAI said on Monday it will start adding an invisible watermark to text written by ChatGPT and Codex for users in the European Union, to comply with the transparency rules of the EU AI Act.
According to TechCrunch, the watermark is not a visible mark. It works by slightly steering which words the model chooses, leaving a statistical pattern that a detector with a secret key can recognise. Because the pattern lives in the words themselves, it stays with the text when it is copied. OpenAI said the watermark does not identify the user and did not noticeably change model performance in its tests.
The feature will roll out over the coming weeks to eligible ChatGPT and Codex users in the EU only. Developers using OpenAI’s API anywhere can switch it on for selected models, but it is off by default. OpenAI also published a technical report on the method, which it calls textGrain.
The company was open about the limits. In one test, swapping 10% of words for synonyms cut detection from about 92% to 66%, and short passages, maths answers and translations are harder to detect. For now, access to the detector will go only to approved researchers and organisations. OpenAI also warned that the absence of a watermark does not prove a human wrote the text.
Why it matters
Schools, publishers and employers want to know when text was produced by AI. This is a step toward that, but the company’s own results show watermarks are easy to weaken, so they should not be treated as proof either way.
Related free tools: Word Counter, Text Diff Checker
Sources
Every ViralScopeHub news item is an original summary written in our own words. We link to the original reporting so you can read the full story from the publisher.