OpenAI to Watermark ChatGPT Outputs by Default in EU
OpenAI will automatically watermark ChatGPT text in the European Union to comply with the EU AI Act, while offering the feature off by default in other regions. The company says its textGrain method is detectable with a key, though such watermarks remain easy to circumvent.
The requirement comes as no completely effective and reliable way exists to watermark AI text. Existing standards such as SynthID and the C2PA project are relatively easy to circumvent for anyone with basic know-how, and OpenAI's watermark is likely to face the same limitation, according to Ars Technica.
OpenAI's method is proprietary and called textGrain. The company has published a technical paper explaining how it works. In general, it operates like other LLM watermarking tools: it places patterns in word choices that are not clear to a human reader and do not meaningfully change the general quality of the output, but that someone with a key can find using a specialized detector.
OpenAI says it will give access to the detector to a limited number of researchers and organizations. It will also provide a request-for-approval process for others to be added over time.