OpenAI begins rolling out text watermarking for provenance
OpenAI says API customers can opt in to text watermarking for selected models, with watermarking planned for eligible ChatGPT and Codex output in the EU over the coming weeks. Its textGrain method adds a statistical signal through model word choices; a detector looks for that signal. OpenAI also plans a controlled research-access program for the detector and says it intends to open-source the technology.
The important caveat is that watermarking is not reliable authorship proof in every setting. OpenAI describes the technology as early and says shorter or constrained passages are harder to detect; strong results in its tests do not guarantee dependable everyday detection. The linked Hacker News item had no readable comments when checked.