Technology & AINews date: Published by Svertok:
OpenAI opens text watermarking to API customers as an EU rollout approaches
The invisible signal may help identify some OpenAI-generated text, but detection weakens with shorter, constrained or edited passages.

OpenAI says API customers worldwide can now opt in to watermarked text output for select models. Eligible ChatGPT and Codex users across all plans in the European Union are due to receive watermarking over the coming weeks, while OpenAI is working with cloud partners to offer it through their services on the same general timetable.
The company is limiting the consumer rollout to the EU at launch so it can learn from real-world use and feedback. Watermarking will not initially become a global default.
What the watermark can—and cannot—show
The system, called textGrain, places an invisible statistical signal in a model's word choices. OpenAI's detector searches for that pattern and reports whether it detects an OpenAI watermark; it does not identify a user or reveal prompts or conversations.
Performance varies with the material being checked. At a target false-positive rate of 1%, OpenAI says the detector found watermarks in about 80% of 200-token psychology passages and about 95% of comparable 400-token passages. Results were substantially weaker for mathematics, where the model has less freedom in choosing words.
Rewriting, translating or editing text can further complicate detection. OpenAI says it is studying those effects and how to distinguish AI assistance from AI authorship, so the result should be treated as one signal rather than conclusive proof about who wrote a passage.
Detector access remains restricted
Approved researchers and expert organizations can apply for detector access, which OpenAI will initially grant case by case to support evaluation and improvement. The company cites the risk of false positives and missed watermarks as the reason it is not releasing the detector publicly at launch.
That restriction applies to text provenance. OpenAI says its verification tools for images and audio, including its web verifier and Content Provenance API, remain publicly accessible to organizations. The company also plans to release the text-watermarking technology as open source, though it gives no release date.
Next

OpenAI added a higher-capacity seat to ChatGPT Business
The August 10 announcement listed Premium at $100 per user per month when billed annually and Standard at $20, with Premium offering five times the usage.
Read next
