What a false positive is
A false positive is an ordinary or acceptable image classified as likely sensitive. Beaches, sports, skin-colored backgrounds, artwork, medical photos, cropped bodies, and screenshots can resemble patterns that a nudity detector associates with sensitive content.
False negatives matter too
A false negative is private content the detector does not suggest. Low light, unusual framing, heavy cropping, clothing, small subjects, illustrations, and video moments between sampled frames can all make detection harder. That is why a scanner cannot certify that a library is clean.
Thresholds trade one error for another
A broad threshold can surface more possible matches but also more ordinary images. A strict threshold can create a shorter list while missing ambiguous content. There is no universal setting that reflects every person’s definition of private.
A safe review policy
- Treat the result list as a queue, not a conclusion.
- Open the full-resolution photo or enough of the video timeline to understand it.
- Check neighboring items from the same moment.
- Keep deletion selections small enough to verify.
- Review Recently Deleted after any cleanup.
- Use a manual pass for dates or contexts where missing an item would matter.
A privacy tool should reduce review work without turning an uncertain model output into an irreversible action.
Privacy and accuracy are different questions
On-device processing answers where classification happens. It does not make a detector perfect. A trustworthy product should explain both its media boundary and its accuracy boundary: where the data goes, whether an account is required, whether results leave the device, and whether the user reviews every action.
How Sensitivity uses results
Sensitivity analyzes media on the device and displays likely matches behind device authentication. It does not upload media for classification and does not delete anything automatically. The user decides whether to keep, hide, or delete each item.