OpenAI to watermark ChatGPT, Codex text in EU under AI Act

OpenAI said on Monday it will embed invisible, machine-readable watermarks into text generated by ChatGPT and Codex for users in the European Union, calling the system textGrain and framing it as compliance with the EU AI Act's transparency rules that took effect on August 2. Gizmodo reported the measure, which starts as an opt-in API feature globally on October 5, 2026 and becomes non-optional for eligible ChatGPT and Codex text output in the EU within weeks.

The watermark works by nudging the model's word choices against a secret key, hiding a statistical pattern that a reader cannot see but a detector can test for. OpenAI says the signal persists through copy and paste, that it "doesn't identify the user, and that it saw no meaningful change in its models' performance." According to TechCrunch, detection drops from about 92% to 66% once roughly 10% of the words are swapped for synonyms, and the technique is weaker on short passages, math answers and translated text. Researchers at the University of Pennsylvania and Yale co-authored the technical report.

For now, only approved researchers and expert organizations get the detector. "These limitations contribute to our decision to provide initial detector access only to approved researchers," the company wrote, and it cautioned that a missing watermark "does not prove human authorship."

Anthropic rolled out comparable text watermarking in August 2026, and Google DeepMind shipped its own version years earlier. The EU AI Act requires AI-generated text to carry a machine-readable identifier; the U.S. has no equivalent federal rule, and OpenAI's new default stops at the EU border.