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.”
i wouldn’t say signature exactly, because that ensures that a video hasn’t been altered in any way: no re-encoded, resized, cropped, trimmed, etc… platforms almost always do some of these things to videos, even if it’s not noticeable to the end-user
there are perceptual hashes, but i’m not sure if they work in a way that covers all those things or if they’re secure hashes. i would assume not
perhaps platforms would read the metadata in a video for a signature and have to serve the video entirely unaltered if it’s there?
You don’t need to bother with cryptographically verifying downstream videos, only the source video needs to be able to be cryptographically verified. That way you have an unedited, untampered cut that can be verified to be factually accurate to the broadcast.
The White House could serve the video themselves if they so wanted to. Just use something similar to PGP for signature validation and voila. Studios can still do all the editing, cutting, etc - it shouldn’t be up to the end user to do the footwork on this, just for the studios to provide a kind of ‘chain of custody’ - they can point to the original verification video for anyone to compare to; in order to make sure alterations are things such as simple cuts, and not anything more than that.
you don’t even need to cryptographically verify in that case because you already have a trusted authority: the whitehouse… of the video is on the whitehouse website, it’s trusted with no cryptography needed
the technical solutions only come into play when you’re trying to modify the video and still accurately show that it’s sourced from something verifiable
heck you could even have a standard where if a video adds a signature to itself, editing software will add the signature of the original, a canonical immutable link to the file, and timestamps for any cuts to the video… that way you (and by you i mean anyone; likely hidden from the user) can load up a video and be able to link to the canonical version to verify
in this case, verification using ML would actually be much easier because you (servers) just download the canonical video, cut it as per the metadata, and compare what’s there to what’s in the current video
Rather that using a hash of the video data, you could just include within the video the timestamp of when it was originally posted, encrypted with the White House’s private key.
It does if you can also verify the date of the file, because the modified file will be newer than the timestamp. An immutable record of when the file was first posted (on, say, YouTube) lets you verify which version is the source.
Apple’s scrapped on-device CSAM scanning was based on perceptual hashes.
The first collision demo breaking them showed up in hours with images that looked glitched. After just a week the newest demos produced flawless images with collisions against known perceptual hash values.
In theory you could create some ML-ish compact learning algorithm and use the compressed model as a perceptual hash, but I’m not convinced this can be secure enough unless it’s allowed to be large enough, as in some % of the original’s file size.
you can definitely produced perceptual hashes that collide, but really you’re not just talking about a collision, you’re talking about a collision that’s also useful in subverting an election, AND that’s been generated using ML which is something that’s still kinda shakey to start with
Perceptual hash collision generators can take arbitrary images and tweak them in invisible ways to make them collide with whichever hash value you want.
