Requiring proof of authenticity is a niche use case, it doesn't seem realistic that this alone will be sufficient to ensure that most cameras in the world gain extreme security to prevent tampering by users, rogue employees in manufacturing companies, etc, etc. Also, we haven't ever been able to make perfectly secure software - an exploit that allows a key to be extracted from the camera is likely, and camera buyers won't care.

A key is just data. Even if the key can't be extracted from common cameras, there's no magic that can prevent from someone making a device with a known key (or falsely register to the magic worldwide-trustworthy central registry that USA, China and Russia trusts but aren't able to influence to get keys for manufacturing fake news) that a particular camera has been made), one that can be used to sign data outside the camera context to make it appear that it was "securely" seen by a legitimate camera.

You just have to make it hard enough that the likelihood of it being tampered with is small enough to be acceptable for a court.

If you're up against the NSA then pretty much any tech evidence would be invalid. But if you're having a small claim court case, then it seems reasonable that authentication technology can keep up with faking technology.

What's to keep you from adding the key to a camera after falsely generating the media? Or using the key of a known camera to generate false footage?

So spitballing, my assumption is that it'd end up looking something like SSL certificate chains today. In the situation you mentioned:

1. The attacker wouldn't have access to the original cameras private key. 2. The attacker creates a fake video, creates a private key and signs the video with the key 3. The attacker tries to install the key into a camera

Step 3 has to be made impossible. Meaning that a camera becomes a trusted entity and only allowed parties (ex the camera manufacturer) has the authority to insert a private key into a camera. This would be that after installing a camera with a new private key, the key would then need to be signed with the manufacturer's private key and also stored in the camera. This shows the manufacturer is responsible for the contents of the video. If someone tries to change the camera's private key it would no longer match what the manufacturer signed.

Which yes then means we need to have authorized camera manufactures and a process for certificate revocation and all of that.

The exact opposite of "free and open" for recording devices - and the only way we aren't flooded with fake videos flooding and seriously impacting society. We have to want this if we want to still have a concept of video evidence in either the justice or social spheres.

This isn't a completely technical problem, there's not going to be an air-tight technical solution.

I think, at this point, anything more than the camera having a secure device-generated key that it uses to sign its output and produce watermarks is overthinking it. That will add a nearly insurmountable barrier many classes of forgery, namely one were a forger claims to have a photo taken with your camera.

You also have to remember a significant class of photo-forgery would involve images that are claimed to originate from a camera the attacker controls. For instance, forged video of a robbery from a security camera. That could, in some cases, be defeated by observing that the images have authentication information they shouldn't have (such as traces of watermarks from other cameras).

At the end of the day, forgeries will be spotted by observing that they have subtle errors that don't add up. That's how it's always been done.

Require that videos submitted as evidence use keypairs which were provably registered or published prior to the event.

Don't have a registered key? Great, but that's no longer admissible.

Still, problems would exist with letting the camera read the private key for signing, but not allowing an attacker (forger) to access the private key and thus sign their modified footage.

Just a thought.

Courts will use the best evidence available for a particular case, including all kinds of things that obviously can be less than perfectly reliable. Physical photographs have been doctored (or staged) for more than a century, and that doesn't mean that photos as such are inadmissible as evidence; so the fact that the process for taking video wasn't especially tamper-proof won't by itself be sufficient to disqualify it.

If a video is available and doesn't contain a pre-published key then the courts will still want to use it as useful evidence instead of simply ignoring it - sure, the opponent might have a slightly easier time attempting to contest that evidence as possibly doctored (just as they can do now), but courts will still make a judgement for each individual case.

Atoms are harder to move around then bits, and the rapid development in our ability to generate convincing video means that it will be hard to predict what sort of evidence we can generate now that will be convincing in the future.

How easy is it to put a different key into an iPhone? How easy is it to get a private key within an iPhone?

Timestamped signatures + publishing the public key w/a timestamp (blockchain?).

I think this is the answer. But then who's going to maintain a worldwide database of all the hashes of all the photos and videos made, and who would we trust to do it?

I suspect the recent Kodak blockchain adoption might be connected with this use case. Not specifically, but with respect to getting into the technology and acquiring IP and expertise.

Canon had a cryptographic add-on module for image verification for their 1D series way back in the 2000s but it was cracked.

I don't know if they subsequently updated it ( was called OSK ) but the idea has been around for some time. It was considered important for submission of evidence and several news agencies also insisted on it.

I think in general, as creating fakes becomes more and more prevalent, society is going to have to view crypto as more and more a necessity instead of as a niche thing.

What if after I'm done creating my fake picture I carefully take a picture of it with an authenticated camera?

with a bit of setup you could remove the lens (while keeping it electrically connected for EXIG purposes) and then project whatever image you want on to the sensor.

DSLRs already do this, but it's been cracked in the past (Canon and Nikon at least).

Wouldn't this limit video post-processing (e.g. color correction, etc.) ?

As long as you keep the original this should be fine. If anyone needs proof just show them.