AI summarized from verified sources
EU text provenance makes AI text easier to identify
It gives teams more ways to check whether text came from AI.
SOURCE CHECK
3 sources
Sources
Key Points
- 1API watermarking is opt-in
- 2EU ChatGPT and Codex are next
- 3Detector access starts limited
- 4Short or edited text is harder
OpenAI outlined how it will meet EU text provenance rules. Customers can opt into watermarking in the API, and EU ChatGPT and Codex text will gain invisible watermarks over time, though short or edited text remains hard to detect.
Key points
OpenAI outlined a plan for stronger text provenance in line with the EU AI Act. Watermarking can be opted into in the API, and EU ChatGPT and Codex text will be covered gradually.
Impact
It becomes easier to check whether a passage was AI-generated. But short passages and edited text are still difficult to detect, so this is not a perfect solution.
What changed
OpenAI outlined how it will meet EU text provenance rules. Customers can opt into watermarking in the API, and EU ChatGPT and Codex text will gain invisible watermarks over time, though short or edited text remains hard to detect.