Our approach to EU text provenance rules
OpenAI explains its phased text watermarking approach to comply with the EU AI Act, addressing limitations and responsible use.Summary
OpenAI outlines its approach to text provenance in response to the EU AI Act. The company is rolling out text watermarking using its textGrain technology: API customers globally can opt in starting now, while watermarking will be added to eligible ChatGPT and Codex outputs in the EU over the coming weeks. Detector access will initially be limited to approved researchers and expert organizations. OpenAI acknowledges significant limitations: detection is less reliable for short or constrained text (e.g., about 80% for 200-token passages versus 95% for 400-token passages at a 1% false positive rate) and can be weakened by editing (replacing 10% of words drops detection from 92% to 66%). Watermarking does not measure human contribution, establish ownership, identify users, or verify accuracy, and its absence does not prove human authorship. The company will continue to refine its layered provenance approach, including Content Credentials, SynthID, and verification tools.
(Source：OpenAI)