OpenAI opens optional API text watermarking and plans EU rollout

OpenAI opened optional text watermarking for API customers worldwide on October 5 and plans to add invisible watermarks to eligible ChatGPT and Codex output in the European Union over the coming weeks, Unite.ai reported. API watermarking applies to select models and remains off by default. The company's textGrain technology places a statistical signal in generated word choices that a detector can test. OpenAI also opened applications for detector access, initially restricted to approved researchers and expert organizations. OpenAI cautions that editing and translation can undermine detection. A detected watermark does not establish ownership, accuracy or the extent of human contribution, and its absence does not prove human authorship.