OpenAI's October 5 text-provenance announcement describes a phased rollout of textGrain watermarking for eligible ChatGPT and Codex output in the European Union. API customers globally can opt in for supported models, with watermarking off by default. Applications for the text detector initially go to approved researchers and expert organizations. This is not an announcement of unrestricted public text detection.
The system embeds a statistical pattern in word choices. OpenAI says short or constrained passages are harder to detect and editing can weaken the signal. Its stated reason for the EU rollout is responding to EU transparency requirements; that is the company's account of its implementation, rather than our finding that a particular user's publication complies with the law.
A signal has a narrower job than a credit
OpenAI's help documentation says a watermark does not establish ownership, accuracy, legal responsibility or the extent of human contribution. A passage may involve a model processing supplied material as well as generating new text. The absence of a detected signal likewise does not prove human authorship.
For an editor commissioning a piece, our practical reading is to keep the production record alongside the finished words. Ask what the writer supplied, which passages were generated or revised, and who checked the claims. A detector cannot reconstruct those decisions from a binary result.
The distinction also matters when an agency delivers campaign copy. An agreement may require disclosure of a tool's use, a particular approval process or retained source material. A statistical signal should not be substituted for the evidence that the team completed its agreed work. Those contractual obligations depend on the actual commission; this article does not determine them.
Avoid the inverse claim
An undetected passage deserves the same source and authorship checks as any other. Treating absence as a certificate would reward precisely the cases in which the tool's coverage is incomplete.
The useful role for provenance is to add context to a wider record. The human decisions about reporting, editing, permission and responsibility still need to be documented by the people who made them.