Dev News Daily ENDE

OpenAI watermarks ChatGPT text in the EU and keeps its detector for approved researchers

OpenAI set out on 5 October how it will meet the EU AI Act's requirement that generated text be identifiable in a machine-readable way. The approach has three parts, and the company is open about the limits of each.

What changes. Over the coming weeks, eligible ChatGPT and Codex text output in the European Union gets an invisible watermark, across all plans and in the EU only. API customers anywhere can opt in to watermarked output for selected models from today; in the API it stays off by default. Approved researchers and expert organisations can apply for access to the detector, which reports whether it finds an OpenAI watermark without identifying the user or revealing prompts. OpenAI says it is not making the detector public at launch because of the risk of false positives and missed watermarks.

How it works. The technology, called textGrain, adds a statistical signal to the model's word choices, and the detector looks for it. OpenAI says it matched or beat other approaches it tested, including Google's SynthID for text, and that it plans to release the technology as open source.

OpenAI watermarks ChatGPT text in the EU and keeps its detector for approved researchers
OpenAI watermarks ChatGPT text in the EU and keeps its detector for approved researchers — Dev News Daily

The numbers that matter. At a target false-positive rate of 1%, the detector found the watermark in about 80% of 200-token passages and about 95% of 400-token passages for content such as psychology, and substantially less for mathematics, where word choice is constrained. Replacing 10% of words with synonyms cut detection on 400-token passages from about 92% to 66%; replacing 25% cut it to 17%.

What a watermark does not show. OpenAI lists it explicitly: not how much a human contributed, not who owns or is responsible for the text, not who the user was, not whether the text is accurate, and an absent watermark does not prove human authorship.

For developers building on the API, the practical point is that watermarking is a per-customer choice outside the EU consumer products, which puts the compliance decision on the integrator.

Written by Victoria Shinder.