OpenAI published its approach to text provenance on October 5, 2026, announcing textGrain, an invisible statistical watermark designed to help identify text generated with OpenAI systems.
What changed
OpenAI says eligible ChatGPT and Codex text generated in the European Union will receive an invisible text watermark as the rollout expands. API customers worldwide can opt in for supported models.
How textGrain works
The system subtly influences the model's statistical word-selection behavior. Across a sufficiently long passage, those choices form a pattern that a detector can search for. It does not add hidden characters, invisible spaces, or unusual punctuation.
What it can and cannot prove
A positive signal can indicate that an OpenAI system generated or processed text. It does not identify the human user, measure human contribution, establish ownership, or prove accuracy. Failure to detect a watermark also does not prove human authorship.
Why developers should care
Text provenance is becoming an application-level concern for publishing, education, enterprise document systems, and compliance workflows. Developers may need to preserve provenance signals while avoiding unsupported claims about authorship.
Limitations
Detection is less reliable for short passages and after substantial editing. OpenAI says detector access will initially be limited to approved researchers and expert organizations.
Practical takeaway: Treat text watermarking as one provenance layer alongside metadata and content credentials, not as a standalone authorship or truth detector.
Source: OpenAI, Our approach to EU text provenance rules, October 5, 2026.