Home / Technology / OpenAI Adds Invisible Watermark to AI Text
OpenAI Adds Invisible Watermark to AI Text
5 Oct
Summary
- Invisible, machine-readable watermark for AI text is now available.
- This feature is launching first for users within the European Union.
- Watermarking aims to meet EU AI Act requirements and doesn't verify accuracy.
An invisible, machine-readable watermark is being introduced to AI-generated text output from ChatGPT and Codex, with an initial rollout exclusively for users in the European Union. OpenAI's textGrain technology is designed to meet the stipulations of the EU's AI Act, a move also being made by other AI developers like Anthropic.
While OpenAI reports that textGrain's performance is comparable to other watermarking methods, it explicitly states that the technology has limitations. The company emphasizes that text watermarks do not ensure reliable detection, nor do they serve to verify the accuracy of the text, establish ownership, measure human contribution, or prove human authorship.
