Saved report copy
Recorded
OpenAI to watermark ChatGPT text and code in EU
OpenAI will add an invisible watermark to eligible ChatGPT and Codex text and code generated in the European Union to comply with EU AI transparency rules. Its system, called textGrain, adds a statistical signal to word choices that a detector can identify without identifying users or revealing prompts, and OpenAI said it matched or exceeded alternatives like SynthID for text but detects short, mathematical or edited text with about 80 percent success. The move follows a similar step by Anthropic and comes ahead of a Dec. 2 deadline under Article 50 of the EU AI Act requiring generative AI providers to make outputs machine-readable as identifiable, with detector access initially limited to approved researchers and expert organizations and the technology planned for open source.