The White House wants to ‘cryptographically verify’ videos of Joe Biden so viewers don’t mistake them for AI deepfakes::Biden’s AI advisor Ben Buchanan said a method of clearly verifying White House releases is “in the works.”

    • drathvedro
      link
      fedilink
      English
      arrow-up
      2
      ·
      edit-2
      5 months ago

      Oh, they’ve actually been developing that! Thanks for the link, I was totally unaware of C2PA thing. Looks like the ball has been very slowly rolling ever since 2019, but now that the Google is on board (they joined just a couple days ago), it might fairly soon be visible/usable by ordinary users.

      Mark my words, though, I’ll bet $100 that everyone’s going to screw it up miserably on their first couple of generations. Camera manufacturers are going to cheap out on electronics, allowing for data substitution somewhere in the pipeline. Every piece of editing software is going to be cracked at least a few times, allowing for fake edits. And production companies will most definitely leak their signing keys. Maybe even Intel/AMD could screw up again big time. But, maybe in a decade or two, given the pace, we’ll get a stable and secure enough solution to become the default, like SSL currently is.

        • drathvedro
          link
          fedilink
          English
          arrow-up
          2
          ·
          5 months ago

          Oh, so Adobe already screwed it up miserably. Thanks, had a good laugh at it

        • Natanael@slrpnk.net
          link
          fedilink
          English
          arrow-up
          1
          ·
          edit-2
          5 months ago

          Oof.

          They need to implement content addressing for “sidecar” signature files (add a hash) both to prevent malleability and to allow independent caches to serve up the metadata for images of interest.

          Also, the whole certificate chain and root of trust issues are still there and completely unaddressed. They really should add various recommendations for default use like not trusting anything by default, only showing a signature exists but treating it unvalidated until the keypair owner has been verified. Accepting a signature just because a CA is involved is terrible, and that being a terrible idea is exactly the whole reason who web browsers dropped support for displaying extended validation certificate metadata (because that extra validation by CAs was still not enough).

          And signature verification should be mandatory for every piece, dropping old signatures should not be allowed and metadata which isn’t correctly signed shouldn’t be displayed. There’s even schemes for compressing multiple signatures into one smaller signature blob so you can do this while saving space!

          And one last detail, they really should use timestamping via “transparency logs” when publishing photos like this to support the provenance claims. When trusted sources uses timestamping line this before publication then it helps verifying “earliest seen” claims.