I saw a news report about now freely available software to make "deepfake" videos. Couldn't videos be internally marked using a private key, so that everyone could verify the originator using a public key? This could be built in to browsers so that everyone could see if something was fake or not.
Is it technically possible to mark a video stream throughout with something that can't be spoofed or removed? Then if a video had no mark, we would know it was rubbish.
In theory, yes. Signing a video file with a private key and then publishing the public key is no different than signing some text and then publishing it.
But this doesn't really solve the problem. For example, imagine someone filmed a video of me putting on two differently colored socks - which, as we all know, is one of the worst imaginable crimes. The person, who shot the video, signs it with their private key and publishes it.
Now I vehemently deny the legitimacy of the video, saying it was obviously faked and I would never wear differently-colored socks. As it turns out, my claim was correct. Someone shot a video of me wearing socks and then modified the video file, to alter the color of one of my socks. He then signed this modified file and published it.
As you see from this example, signing a video really doesn't "verify" that the content of the video is "legitimate" in one way or another.
it makes things even worse. "Deep Fakes" are primarily a social problem, meaning that a significant amount of people believe that the content of the video is real, despite it not being so. You cannot fix social problems with technology, as that tends to create even more social problems.
By adding a simple green checkmark of "Legitimate" to a video, you essentially teach people not to engage their brains and question what they see, and instead create a shortcut of "checkmark = truth". And while some people might not be fooled by it, keep in mind that propaganda doesn't need to work on everyone, just enough people.
In short: The best way to combat deep fakes is to teach people to think critically.
Yes, you can use cryptography to authenticate video sources.
No, you cannot use cryptography to prove a video is authentic (and prevent deepfakes).
If these statements sound contradictory, read on.
This has been done in the past by camera manufacturers like Nikon and Canon, who wanted to assure pictures taken with the cameras could be "authenticated".
The Nikon Image Authentication System was quickly broken by a simple firmware dump attack.
It's important to note that virtually all known schemes that attempted to provide security against a physical attacker have failed so far (sometimes in secret, for a while). The general saying is that "If an attacker has physical access to my machine, all bets are off".
If someone developed a perfectly secure system that resisted attacks for many years (with the help of tamper-proof chips / TPMs), it would only ever be able to verify something very narrow: that the produced image/video was indeed processed by a certain company/manufacturer, or certified as true from a certain government / news source
This is useful for ensuring provenance (who says what?) in some controlled scenarios, but doesn't help at all for ensuring authenticity (what is real?) of the content.
Things this system doesn't prevent include:
So let's say these limitations don't bother you and you want to go ahead and build the system anyway. After all, no system is perfect, and something is better than nothing, right?
Once you create this system that certifies what is real, several things are likely to happen:
tl;dr: Cryptography is useful, but it can't replace trust. Centralizing decisions about trustworthiness leads to poor outcomes
This is analogous to how big media companies and some governments "think" - there is good content, and there is bad content. The good content is somehow marked as good, anything else is bad. Good people control the "good" mark. All problems solved.
Why the evil bit-type solutions don't work is more or less widely known.
Why the ministry of truth-type solutions are risky and of limited use is also known.
Besides, the deepfake technology is just one more technology. We have cinema industry for a good century now. We have makeup experts that can paint anyone's face over an actor. Deepfake is not any better except maybe being cheaper and faster, to an extent.
On the other hand, the whole deepfake drama in the recent times stems from the possibility to publish an information anonymously. This is an important feature of more or less free societies and an important tool for keeping them free. In regrad to this, an occasional deepfake now and then are not that much of a high price.
What you propose will not kill deepfakes, it will kill anonymity. This is pretty much not the same.
This is not to say that digital signing of a camera footage is not legitimately used for other purposes. Security cameras, dash cameras, professional reporter cameras do this (as well as extensive digital signature-based timestamping). This (especially the third-party timestamping) in some cases can be used to prove authenticity.
Think about the private key distribution for a minute. Who do you think should be able to sign videos?
users registering freely. This doesn't give videos any additional trustworthiness: why would I trust a video signed by [email protected]?
users which confirm their identity. This would make things worse: users posting videos would lose anonymity. People will be intimidated to post fakes, but they will also be intimidated to post controversial or anti-government content. Plus, I still won't be able to decide with reasonable certainty whether to trust a video from John Doe, 42 Random str., Metropolis or not.
certified recording equipment. This would make fake videos somewhat harder to produce: a fake video maker will not be able to use just a video editor. You still won't be able to tell if the video was staged or not, whether it really displays the place and the people it claims to represent, etc. Plus, you'll automatically mistrust videos from users who don't have the certified equipment, which are not necessarily fakes.
certified organizations. Those don't need to sign their videos at all: they can simply publish their videos on their own website, and unless their website is cracked you can be sure those videos are made by them. If they decided to post a fake, they'll have to problem to forge the signature as well.
An of course, all these options still allow for scenarios where the private key gets compromised, or trusted parties become malicious under government pressure or criminal threats. Discussing these scenarios only makes sense if there is an option that works.
I agree with the others that have opined on this question that a signature on the video in itself does little to authenticate the video. What matters is who made the signature.
In that regard, it's similar to the system of PKI that we use to authenticate SSL certificates on the web. Our web browsers do not trust just any signature on an SSL certificate - our browsers only trust signatures made by certificate authorities (CA's) that our browsers trust. We put our trust in these CA's to authenticate certificates on the web, and if the certificate has a valid signature by a trusted CA, then we feel confident that the certificate is authentic.
Perhaps we could we have a similar ecosystem to authenticate videos, using 'video authorities' (VA's ?) that we trust. If I trust @EasyWhenUknowHow as one of my trusted VA's, and the video has a valid signature by @EasyWhenUknowHow, which I've verified using EasyWhenUknowHow's public key, then I can feel confident that the video is authentic. I smell a business opportunity...
Any person or organization X can sign any data (a video or otherwise) attesting various things, including...
But the signature can be applied whether or not the claim is true or false. So, the signature by itself does not prove if the content is real or not. It only proves that someone claims it is, and possibly who is making the claim.
So how do we use that?
If a party wants to deny any involvement with some content, then they simply can refuse to sign it. That is an open problem. But conversely, if a video featuring person X is signed by person X, then we can at least know that X is claiming it's really them.
With respect to "Deep Fakes" there is little practical difference between...
In both cases the signature itself proves that X claims Y. One can't know if X is lying, but that would also be true if you were in the room with them as they said it.
In summary we can't know if a video of X is fake if X claims its fake. But we can know if X claims it's true.
Then if a video had no mark, we would know it was rubbish.
That's an interesting proposition. A lack of a signature doesn't prove the content is fake, but it does prove that person X did not choose to claim it is true.
If person X had a policy of always signing any official content, and the content is merely claiming that X has opinion Y, then a lack of a signature by X is a pretty strong indicator that X doesn't claim to have that opinion.
Often times we get content (videos included) from trusted websites. But the content often gets copied and lives on long after it's taken down from the website. A signature can prove that that a specific website at one time claimed to have hosted it.
Public figures often proudly put out content only to later deny its existence.
Just knowing that someone claims something can actually be useful by itself, whether or not the claim is true. For example, politician X wants to unequivocally pronounce their support of popular cause Y by signing some content and putting it on their website. 10 years later, cause Y is really unpopular, and the content is removed from their website.
Now politician X want to deny that they ever really supported cause Y, but their opponent digs a signed copy out of an internet archive (proving that X is now either lying or really did support Y).
A shrewd politician manages to talk a lot without every really saying anything, and certainly never digitally signs something. But not all of them are that wise, and this use-case would probably occur quite a lot.
In general, we can't know for sure if the content of a video is true, but we might try to assess the probability of veracity, based on how much we trust the signer. It's not perfect, but it's a useful heuristic that humans use all the time.
Yes, this is possible and there is already an initiative to implement this called The Coalition for Content Provenance and Authenticity (C2PA) containing major industry partners.
From their summary, trusted hardware will digitally sign information about a media object (image or video) when it's created, and as the object is modified those transformations are also signed.
There are two very different points here.
Can we sign digital videos:
Obviously yes, a video file is nothing but a file and it is possible to take a hash of it and sign it. From that point there is an unbreakable bond between the signed file and the signer
What will that guarantee:
The video will deserve exactly the same trust that the signature does. If a well known press company signs a document (whatever the content) and you do trust that company, then you can trust the document. If somebody you do not know signs the document with a reputable non-repudiation signature emitted by a reputable CA, that means that an individual human being will be liable for the content of the file and that legal actions could be used is it was a fake - well if all that occured in a reputable country... Less strong that the first use case, but the fear of legal actions is usually enough to calm down many people. If the signature has no legal value (or only little one depending of the country it was emitted in), it just means that the document has not been modified since it was produced. Nothing less nothing more. Specifically it makes no evidence that it is or not a fake.
An interesting outcome of this is that you cannot full rely on a browser for that. You can find certificates emitted by reputable CA for Chinese or Russion citizens. As far as I am concerned, I would not trust a video about Ukraine war if it was by a Russian organization or citizen. Because I would not trust the Russian legal courts to have the same sight on truth that I would have...
When signing code packages, one of the things we often have to obtain is a signed timestamp from Digicert or someone. If the timestamp could be obtained from a reputable source and cryptographically linked with the video, you would have a basis to prove that the video was generated at the purported time of the event, where a deepfake would likely take a significant amount of time to generate.
That doesn't solve everything, of course. You could have a pre-trained deepfake model ready to go, or you might not be able to prove the time of the event in the video, but it would go a long way to proving authenticity. I do think MechMK1's answer makes an important point in saying that trust and opinion often matter more than proof.
You can digitally sign perceptual hashes of images or frames in videos like proteus.photos does, allowing you to reverse search those images even when posted online. You can think of it as C2PA + private reverse image search.
