So the way the lawyers are approaching this evidence is their usual cavalier approach to truth. They are not attempting to establish “my client never said that,” but just trying to poke holes, “this video of my client saying that is not sufficient proof that my client said that.”

There is for example no statement under oath from Elon, “that video was a fake, that absolutely never happened.” Even without that, there is no paper trail of “my client maintained a very consistent messaging that this was not fully autonomous driving and you need to always be vigilant... this isolated clip is so strange that we need to seriously entertain the idea that it was deepfaked.” That sort of stuff.

Regarding the faithful wife accused of cheating, I mean, there are lots of valid ways to investigate. First questions are “where did this video come from, where do you believe it was filmed, why do you believe it is your wife rather than someone who happens to look a lot like her,” etc. ... Normally a PI would be claiming “I followed her to so-and-so motel, I took this video legally from outside” and even then you might get objections from a decent divorce lawyer, “this is a two-party consent state and my client didn't consent, this is unlawful surveillance,” yadda yadda yadda. But for a deepfake you have all those objections and more.

"I received it in an email from a guy saying he is sleeping with her. She says her name and everything"

Look up “laying the groundwork” for “documentary evidence” and you will find that this fails to show chain of custody unless you can get that “guy saying he is sleeping with her” in the courtroom.

This isn't a new problem with video, but a new twist on an old problem: manufactured evidence.

One way that courts deal with it is enhanced scrutiny on disputed evidence. Like, if you show up to a court with a document that you claimed someone signed, and they claim they didn't sign it, then the court can keep digging until they figure out who is lying. They can look for witnesses. They can subpoena records from internet companies that may have been involved in a digital transmission of that document. And so on. It's still possibly to lie to courts and manufacture evidence, but most people are not going to cover their tracks well enough to do so without getting caught if the full force the court's scrutiny is turned on them.

And the court finding out you lied to it is very very bad for whatever case you were trying to make.

perjury isn't looked to kindly on in the courts.

I've heard this is hardly ever prosecuted except in federal court

> if the full force the court's scrutiny is turned on them

good luck doing that without severely deep pockets

Everything will be considered fake unless it is proven authentic.

How do we prove authentic? Same way we prove that a website we visit is authentic. Every recording made will be signed/hashed and be able to validate of the public key of the recording device (every phone will have a key). White House, News Orgs, etc. all will use cameras with public keys so everyone can validate. As soon as only one pixel or frequency changes, it will be invalid/fake.

In the end, all will be even more trustworthy than today (there will be some kind of signed/verification label on every video/audio/media that is original/valid).

That would represent a revolution in terms of how the court operates and how many cases are thrown out. There will be riots.

You could take a picture of a picture or a recording of a recording and now it is validated.

But it wouldn't be validated with the original key, so people could know it could have been tampered with.

If the key is just the recording device, why not? Play a fake on some other device and record with your own.

Datestamps are part of the keyed metadata.

My camera has no clue what date it is beyond what I tell it the date is.

Which makes video evidence from your camera more suspect than video evidence from a cellphone (which, sure, could be fake-dated as well).

Which further reinforces the fact that there is a tremendous amount of real evidence that will be thrown out.

The idea is that the blockchain tracks the date the video was entered onto the blockchain verifiably. Note that this isn't the same as the date the images were taken, but it confirms that the recording is at least X days old.

How would that relate to the hashing explained here?

But then the metadata, e.g. gps coords, time of the recording, etc. of the recorded video will not match with what is displayed. So it would be an obvious fake.

GPS can be spoofed by replay

You know, contracts are just pieces of paper. They can easily be faked. Yet try going on court and say that a contract with your signature is invalid because it could be fake. Your argument will be rejected, because it doesn't matter if it could be faked, it matters if you are alleging that it was faked.

So the same thing should be the standard for video. Someone shows a video of you doing something. Either say under penalty of perjury, that you did not do that, and then the authenticity of the video can be looked into, or the video will be accepted as authentic. You don't get a preemptive "this could be fake, so lets ignore it".

This is a good idea and would help a lot (though not a perfect solution with the whole "picture of a picture" thing).

I wonder if there is any traction in the industry towards making this happen?

It's not a new problem. There was an HN article I was reading about someone facing police audio tape evidence and their defence lawyer said it must have been edited to make their client look guilty. The prosecution proved the recording was continuous and not edited from the pattern of mains hum noise on the tape. There'll probably be some clever technical methods to differentiate deep fakes from original recording along soon. Then there's just simply the weight of other evidence that may be present e.g. whether witnesses agree they said the statement etc.

I don't think the same will apply here. Even original record methods are dubious but for deepfakes it's just a matter of time to when it's completely indistinguishable from the real thing

> I'm surprised people aren't using this more for divorces and what not. Create a deepfake of your wife and then say "Here's proof my wife was cheating"

No fault divorce makes it kind of pointless to use in divorces.

Are all states no fault states? Also I think it can matter for prenups. Cheating can equal no money

Videos are easy enough to spot as fakes for now, but I've heard really convincing AI audio fakes. Audio only recordings might not be usable as evidence without additional confirmation of authenticity.

What would "additional confirmation" look like?

Are audio/video recordings I take with my phone going to be useless in a decade because deepfakes are virtually indistinguishable and there is no "additional confirmation"?

That is a good question, because there are two trends at play here. On the one hand, you have ever improving ability to create deepfake images, videos or audio recordings. On the other hand, AI-based retouching of photos and videos is now pretty much standard on mobile, some of it done by default and hidden from you in software. It wouldn't surprise me if the same was case with voice too.

Point being, it'll take more than just looking at the source image/video/audio for signs of AI manipulation, because soon enough, all recordings will be AI-manipulated before even hitting persistent storage.

In other words: everything will be deepfake by default, and the issue will turn into telling which deepfakes were done to lie about substantial aspects of reality pertinent to the case, and which ones were made to enhance some non-relevant qualities (aesthetics, compressibility, etc.).

Useless? No. Possibly open to being faked? Yes.

Lots of ways to sign photos and videos today that should make them fully verifiable in the future (assuming public/private key cryptography remains secure). For stuff that isn't signed, it's not like good old detective work can't verify in a lot of cases. Phone records, eye witnesses, additional unpublished recordings, etc.

Also being verified really only matters when it comes to "truth" as defined by law. It's not like your family won't enjoy your videos of Christmas '23.

a cryptographically-verified video of a phone taking a video of a screen playing another (fake) video, is not very useful though.

Additional confirmation could simply be multiple sources reporting the same thing. Once is circumstantial, but if 10 people record something from different perspectives then it is highly likely to be real.

It is 100% not an issue. The human eye and ears can’t tell if some video or audio is fake, but you cannot hide from the machine. Deep technical analysis will easily reveal something is manufactured, as easily as it can tell an image is photoshopped.

And God help you if they discover you manufactured evidence. Manufacturing evidence raises the stakes considerably. It makes any crime instantly more heinous, proves you deliberately had bad intention. You might go in thinking you’ll win easily with manufacturerd evidence and come out with greater felonies and complete eradication of your reputation.

You think you might be so fucking smart using off the shelf components to make yourself a deep fake, rest assured there are always people smarter than you out there. Intellects vast and deep that will easily pierce through your lies and destroy you. Do not attempt to provoke them. They will relish in picking your remains apart.

Human eyes and ears can tell if it's fake? Lol

This is a ridiculous comment

I do wonder if for areas where provable authenticity is necessary we will see a move back to analogue film.