OpenAI plans EU textGrain watermarks for ChatGPT and Codex
Eligible EU users will receive invisible text watermarks over the coming weeks; OpenAI says detection has important limits.
Regional rollout and optional API use
OpenAI is adding an invisible, machine-readable watermark called textGrain to text generated in ChatGPT and Codex, The Verge reported on October 5. Quoting OpenAI’s announcement, the publication says eligible users across all plans in the European Union will receive watermarking over the coming weeks. OpenAI is not making it the global default at launch. API customers worldwide can opt in to watermarked output for select models, while the company works with cloud partners on availability through their services.
A provenance signal with limits
The Verge reports that approved researchers and expert organizations may apply for access to a detector. OpenAI says access will be granted case by case initially, rather than opening the tool to everyone. The company acknowledges both missed watermarks and false positives, so a detector result is not definitive proof of where text came from. OpenAI also says a watermark cannot verify factual accuracy, identify ownership, measure a person’s contribution or prove human authorship. The feature addresses a transparency need under EU rules, but its real-world reliability remains to be tested during rollout.