OpenAI said on October 5, 2026, that it is beginning a limited rollout of textGrain, an invisible statistical watermark for AI-generated text. API customers worldwide can opt in for selected models; the setting will remain off by default. Over the coming weeks, eligible ChatGPT and Codex output for users in the European Union will also carry the mark.
The system slightly steers a model’s word choices to add a pattern that a detector can test for later. The signal is woven into how text is generated rather than shown as a visible label. OpenAI is accepting applications for detector access from approved researchers and expert organizations, but is not opening it to the public while it evaluates missed marks and false alarms.
OpenAI’s reported tests show why detection depends on the text. A token is a small unit a language model uses to process text. At a target false-positive rate of 1%—the rate at which unwatermarked passages would be wrongly flagged—OpenAI says its detector found the watermark in about 80% of 200-token psychology passages and 95% of 400-token passages. Results were substantially lower for mathematics, where there is less freedom to vary wording. In a separate test on 400-token passages, replacing 10% of words with synonyms lowered detection from about 92% to 66%; replacing 25% lowered it to 17%. These are company evaluations, not independent field results.
A positive result is limited: it indicates a likely OpenAI watermark, not who wrote the passage, how much a person contributed or whether its claims are true. A missing mark does not prove human authorship; short, translated or edited text can lose the signal.
The rollout follows the August 2, 2026, start of transparency obligations under Article 50 of the EU AI Act. The European Commission says provider duties include machine-readable marking and detectability as far as technically feasible; its code of practice is voluntary, while the law’s requirements are not. OpenAI’s announcement describes a phased technical rollout and does not by itself establish legal compliance.
Sources
- OpenAI: Text-watermark rollout and evaluation
- OpenAI technical report on textGrain
- European Commission: Code of Practice on Transparency of AI-Generated Content
Image credit: ResDigital18, Wikimedia Commons, CC BY-SA 4.0. Illustrative photograph.
