OpenAI to watermark ChatGPT and Codex text in the EU
OpenAI laid out how it will comply with Article 50 of the EU AI Act: over the coming weeks it will automatically add an invisible watermark to eligible ChatGPT and Codex text output in the EU across all plans, while API customers worldwide can opt in for select models starting today (off by default). The method, called textGrain, subtly shapes the model's word choices to embed a statistical signal that a detector can read; detector access will initially be limited to approved researchers and expert organizations, and OpenAI plans to open-source the technology. OpenAI's evaluations put detection at about 80% for 200-token passages and about 95% for 400-token passages at a 1% false-positive rate, lower for content like mathematics, and replacing 10% of words with synonyms cut detection from about 92% to 66%. The company said it saw no meaningful benchmark differences for its latest frontier model, Astra, with watermarking on.
Text provenance is becoming a compliance requirement in the EU, but the technology's limited reliability means a watermark—or its absence—can't establish authorship.