• phillaholic@lemm.ee
    link
    fedilink
    English
    arrow-up
    4
    ·
    1 year ago

    They published a white paper on it. It would have taken many detected examples before they did anything about it. It’s not strictly a hash as it’s not looking for exact copies but similar ones. Collisions have been proven, but afaik they are all reverse engineered. Just Grey blobs of nonsense that match CSAM examples. I don’t recall hearing about someone’s random taken photo matching with anything, but correct me if I’m wrong.

    • Kelsenellenelvial@lemmy.ca
      link
      fedilink
      English
      arrow-up
      2
      ·
      1 year ago

      True, it’s hash-like in that the comparison is using some mathematic representation of the source material. It was intended to be a little fuzzy so it would still catch minor alterations like cropping, watermarks, rendering to a new format, etc…

      The example I heard of was someone that was using an app for a remote doctors appointment. The doctor requested photos of the issue, a rash in the genital area of a minor, supposedly one included an adult hand touching the area involved. That photo ended up in Google’s cloud service where it was flagged, reported to law enforcement, and that users while Google account was frozen. The investigation quickly confirmed the innocence of the photo, and provided official documentation of such, but last I heard Google would not release the account.

      • phillaholic@lemm.ee
        link
        fedilink
        English
        arrow-up
        1
        ·
        1 year ago

        Google has unencrypted access to your files to do whatever they want with, do we know this was the same CSAM system or one of Google internal ones? Google Photos does their face and object scanning on the cloud where apple does it on device.