OpenAI is adding text watermarking in ChatGPT and Codex

OpenAI is introducing an invisible, machine-readable watermark called textGrain in the text output of ChatGPT and Codex. The feature will initially roll out only to users in the European Union, across all plans, over the coming weeks; it will not be a global default at launch. The regional approach is intended to allow OpenAI to learn from real-world use and feedback before broader deployment. OpenAI claims textGrain matched or exceeded other approaches such as Google DeepMind's SynthID for text, which also underlies the watermarking Anthropic announced in August; Anthropic likewise acted to meet EU AI Act requirements.

OpenAI published AI benchmark scores showing similar performance between watermarked and unwatermarked text, but cautions that textGrain does not guarantee reliable detection and that text watermarks do not verify accuracy, determine text ownership, measure human contribution, or prove human authorship. In addition to the EU rollout, API customers worldwide can opt in to watermarked outputs for select models starting today, letting them decide how watermarking fits their transparency obligations. OpenAI is also working with cloud partners to make watermarking available for OpenAI model outputs accessed through their services in the coming weeks.

Approved researchers and expert organizations can apply starting today for access to a detector tool. In line with the Code of Practice, access will initially be granted case by case to support evaluation and improvement of text provenance. The tool will report whether it detects an OpenAI watermark without identifying the user or revealing their prompts or conversations. Because of the risk of missed watermarks and false positives, OpenAI is not making the detector publicly available at launch.

(Source：The Verge)