Legal
The detection accuracy disclaimer
Verdicts, scores and notes are opinions generated by software about content. They are not facts about any person, are not verified by a human, and can be wrong in both directions.
Updated
Section 5.2 of the terms is written in capitals for a reason. In plain words:
What a verdict is
The output of an automated pipeline for a piece of content: a label (human, uncertain or ai), a score from 0 to 1, and notes. It is generated by software, not verified by a person.
What it is not
A statement of fact about any person. Proof that someone did or did not use AI. Something to rely on for a decision about a person.
Why it can be wrong
Detectors depend on signals that may be absent, stripped, spoofed or unavailable, such as watermarks and Content Credentials. Classifiers and writing patterns are probabilities. The verdict cache contains rechecks requested by other users and reports submitted by other users, neither of which we verify. See Why false positives happen.
What we promise
Nothing about the accuracy, completeness, reliability or timeliness of any verdict. The service is provided as is, and the terms limit our liability for damages arising from a verdict that is wrong, from content that was hidden or not hidden, and from any action taken in reliance on a verdict. Some jurisdictions do not allow certain exclusions, in which case they apply to the minimum extent the law requires.
What we ask of you
Treat a flag as evidence to weigh, not a conclusion. Use Not AI when it is wrong. Never use it against a person; see Acceptable use.