Main Topic — Apple Reference Image & C2PA

Note: the main topic was recorded as a conversation with Allison Sheridan from the NosillaCast podcast.

We’ve discussed the Content Credentials image provenance standard developed by the C2PA, or Coalition for Content Provenance and Authenticity, a few times on the show over the past few years. I described how it works in LTP 125, and explained why I think the concept of image provenance is so important in the age of AI in LTP 155. I’ve been a strong advocate for C2PA Content Credentials since I first learned about them. But, much to my surprise, Apple have forced me to reconsider their place in our modern AI-infused world, by releasing a very different technology they’ve named Reference Image.

Does Apple’s technology obviate the need for C2PA, or should photographers eschew Apple’s proprietary approach in favour of the preexisting open standard? Should photographers use C2PA, or Apple’s Reference Image?

Like with so much in this world, the answer I’ve come to is ‘yes’. The technologies are similar, and they’re both necessitated by the rise in AI, but they approach the same problem in very different ways, and that really matters. In some situations, C2PA has clear advantages over Apple’s system, and in others, Apple’s system is the only viable option!

A Quick Refresher on Content Credentials

Content Credentials provide a cryptographically verifiable step-by-step history of an image. Ideally that history, or provenance starts from the moment the shutter fires, to continues up to the moment you view the image on your screen. Every edit is captured and catalogued, so you can appropriately contextualise the image.

If it’s a piece of fine art photography, then dramatic edits in something like Lightroom followed by more edits in other tools are utterly unremarkable — that’s all part of the art and craft of creative photography!

On the other hand, a journalistic image with anything more than a few minor exposure fixes and a small crop is utterly untrustworthy!

So, content credentials can provide an important source of information for contextualising images. But of course, we need to be careful not to over-rely on them. A genuine image that has not been manipulated can be extremely misleading by dint of what the photographer chose not to show! An easy example to illustrate this point is images of the same political demonstration shot by partisan photographers with differing agendas. Where you stand, and how wide you shoot can make the same crowd look large an imposing, or small and pathetic, and that doesn’t requite any kind of digital trickery!

To be able to provide us this powerful source of information, Content Credentials build on top of the same Public Key Infrastructure that secures much of the internet. The same X.509 certificates that allow your browser to cryptographically prove you really are on Amazon’s actual website, and not some hacker’s imitation, are used as trust anchors for Content Credentials.

Camera makers use certificates to tie specific cameras to their factories, news organisations use certificates to tie specific cameras to their photographers, and software vendors use digital certificates to tie specific pieces of software to their organisations. The end result is an image you can prove was taken with a specific camera, owned by a specific organisation, that has had a specific set of edits applied in specific order with pieces of software.

That’s really quite something!

An Overview of the Public Key Infrastructure (PKI)

C2PA is all about cryptographically binding media to identities, but what does that mean? How does that work?

A fundamental concept underpinning all of the PKI is a special type of encryption known as asymmetric encryption.

With traditional symmetric encryption there’s just one key, and you use that same key to encrypt and decrypt your messages. Asymmetric encryption replaces that single key with a pair of keys; a public key, and a private key. Anything encrypted with the the private key can only be decrypted with the public key, and vice versa.

This is paired with a second fundamental concept, so-called cryptographic digests, or cryptographic hashes. These are functions that convert any original input to a fixed-size random-appearing output that can’t be reversed. Every time you digest the same data, you get the same short series of random gibberish. Not only is it impossible to take this random gibberish and recover the original information, it’s also not possible to edit the original information a little so it produces the same random gibberish. Small changes in the input result in big changes in the output.

These two concepts are combined to create digital signatures — you take the information you want to sign, digest it, encrypt the short digest with your private key, and that’s your signature. Anyone can verify the signature by decrypting it with your public key, and comparing the result to their own fresh calculation of the digest. If the digest calculated by the verifier matches the digest decrypted with the public key, then that guarantees two things:

1. That the information really was signed by the owner of the matching private key
2. That the information has not been altered in any way since it was signed

These kinds of digital signatures are used to verify each step of an image’s journey in its C2PA provenance.

However, that still leaves us with a problem — how do we prove ownership of the key pairs used to sign each step of the history? Anyone can generate a fresh a key pair in seconds, so can’t they just use theirs and say it’s mine? If it were possible to fake key ownership, then digital signatures would be meaningless!

This is where digital certificates come into play.

The so-called trust anchors of the entire PKI are Certificate Authorities — these are heavily audited organisations that rigorously follow very prescriptive rules for verifying identities, and they then cryptographically vouch for the identities of the key owners by issuing Digital Certificates.

The crown jewels of certificate authorities are their private keys. It’s those private keys that grant them their trusted status, because it’s their public keys that get installed into every computer, phone, and tablet operating system as trusted public keys. This means that every computing device knows the correct public key matching every certificate authority’s private key.

So how do certificate authorities use their position of trust to vouch for key-pair ownership? I want to get a digital certificate to sign things, how does that work?

Anyone who wants to be issued a digital certificate for any purpose, including C2PA, starts by generating a key pair. The private key is kept secret, and the public key needs to be somehow cryptographically bound to your identity by a certificate authority.

This is how that works:

1. You insert your public key and your identifying information into a standard template known as a certificate signing request
2. You submit that request to your certificate authority of choice, and complete their validation process
3. When they’re confident your identity is real, the certificate authority uses their private key to digitally sign your certificate signing request. That signed request, which contains both your public key, and, your identity, is what we call a digital certificate.

So, if I have a certificate, it cryptographically binds my identity to my public key, and I can use that certificate to digitally sign anything I want, binding the media not just to my public key, but to my identity as well! If I send you a digitally signed PDF, you can be absolutely certain what you received came from me, and has not changed in any way since I signed it!

All this leads to one final question: what counts as an identity?

The X.509 certificate standard that powers the PKI defines the types of identity that can be certified. I won’t bore you with the full list, but it includes:

1. Email addresses
2. Domain names (in other words, websites)
3. Postal addresses
4. Registered corporate identities like governments, government institutions, charities, corporations, and sole traders.

That means that large media organisations can use C2PA to bind their work to their corporate identity, and creators can use C2PA to bind their work to their website.

C2PA without Certificates?

Applying for a certificate involves some faffing about, and often, a cost.

Proving ownership of an email address is about as easy as it gets — the certificate authority simply emails some random gibberish to your email address, and if you can copy and paste it back into the form, they issue your certificate.

Proving ownership of a domain name is not much more difficult — the certificate authority gives you some random gibberish and you either publish it to a specially named file on your website, or add it to a special kind of DNS record on your domain.

Proving your corporate identity — that’s where the faffing about really takes off. I’ve been through the process a few times with my work hat on, and it’s not the kind of thing you get done in an hour, it’s a lot more involved, and it requires the involvement of the organisation’s senior officers.

The BBC have no problem proving their corporate identity and getting certs for their photojournalists, and creators can easily get certs for their websites. But typical home users who just snap some shots on their phone will never do any of that.

So how can C2PA work for regular users on their Android phones and in popular apps like Adobe’s creative suite?

The answer is delegated certificates.

Organisations that have certificates can use their certificates to issue certificates of their own. These certificates don’t carry the same weight as those from certificate authorities, but they allow organisations like Adobe to offer users at least some access to C2PA’s potential power.

For example, in the settings for my copy of Lightroom, I’ve enabled C2PA without ever applying for a certificate. I simply entered my details into the appropriate settings fields. What happened under the hood is that Lightroom created and a key pair, and Adobe itself issued a certificate for the matching public key. That certificate does not prove the information I entered is correct, but it does add some C2PA protection to my images.

The provenance on my edited images is incomplete — the verified trail doesn’t start in my camera, and my work in Lightroom is not strongly bound to my identity, but it does still give me some value.

Because neither my old DSLRs nor my iPhone support C2PA, my original images will never be cryptographically tied to me or my camera. But the image, as it was when I imported it into Lightroom, becomes the first cryptographically verified step in the image’s provenance. Every edit I make in Lightroom then gets added to that provenance, and it is cryptographically bound to Lightroom, and my identity claims are cryptographically bound to the image at each stage of its journey.

When I publish such an image, people can verify that:

1. Someone using Lightroom who chose to enter my website into the settings had a copy of the unedited image
2. The details of every edit made to the image from that point forward

This is useful to me because it supports my claim that the image is mine. It doesn’t prove I took it, or that it’s not AI-generated, but it does prove that I had the original unedited version, and that at the time the edits were made, they were attributed to me.

If someone claims I stole the image, but can’t produce a copy of the unedited original, then their claim is not credible.

But equally, the identity claim is weak, because Adobe are not a PKI certificate authority, so they didn’t make me prove the veracity of the identity information I configured in Lightroom. However, my C2PA claim is the digital equivalent of the old practice of physically writing your own name on your negatives, or signing your name on a painting.

This limited C2PA protection is better than no protection at all, which is why I’ve chosen to enable the feature in Lightroom.

What this kind of limited use of C2PA by non-professionals doesn’t help solve is our societal real -v- fake problem.

For C2PA to prove an image is real, you need certificates issued by recognised certificate authorities installed into cameras with built-in C2PA support, and for every edit to those images to be made using tools that support C2PA.

Realistic for major news organisations, but not for regular users.

Apple’s New Reference Image System

And this is where Apple’s new Reference Image technology fits into the narrative!

Apple have not tried to implement a full provenance system, or, to provide any kind of image attribution. Instead, Apple built a system designed to prove that an image was not artificially generated.

Apple’s approach combines a dedicated security chip built directly into the image sensors on the latest pro-level iPhones with secure software running in their Private Cloud Compute infrastructure to create a kind of digitally signed negative that remains attached to the image. It’s this digitally signed negative that Apple refer to as the reference image.

To see how much or how little an image has been altered since it was originally shot, you compare the current image to the digitally signed negative. If the only differences are acceptable tweaks to exposure, then the image is not manipulated, but if you see dramatic changes, like the removal or addition of people or things, then you know the image has been interfered with.

This combination of hardware and software, both on-device and in the cloud, comes with some drawbacks — so this is a mode you need to explicitly enable when you want to shoot a verifiable image. Once you do, you lose access to all but the primary lens, at least for now, because the iPhones 18 Pro only have the needed hardware on the sensor behind their main lenses. You also have to wait a few seconds for each photo to be ‘developed’, because a low-resolution version of the image has to go up to Private Cloud Compute to be digitally signed before it appears in the Photos app.

By design, Apple’s Reference Image doesn’t bind photos to specific cameras, specific people, or specific organisations. It also doesn’t capture details of any of the post-processing applied to the image. Instead, the digitally signed snapshot of the original image cryptographically verifies the following:

1. That the image came from an iPhone camera
2. What the image looked like before a single piece of software had access to it
3. That the image was taken no earlier than a specific point in time

Doing all that is really difficult! Apple describe the full process in a technical description of the feature, but that’s more than we need here in a photography podcast!

Fundamentally, Apple have engineered a system for doing just one single thing — proving an image is not AI generated, and doing so without sacrificing any user privacy.

Not only can you not tell which iPhone a reference image came from, but you can’t even determine if any two reference images come from the same phone. And it definitely can’t tell you who took the photo.

The reference image also allows viewers to see the cumulative effect of all the changes made to an image since it was captured by comparing the image as it is now to the reference image, regardless of how those changes were made.

Compare & Contrast

The way I see it, Apple’s Reference Image fills in the gaps C2PA can’t cover.

If you are someone like Reuters, you want to prove not just that an image is real, but that it was taken by one of your photographers. You want to prove both authenticity and attribution. That’s what C2PA does.

But, if you’re a whistleblower or an undercover journalist trying to document some atrocity in some dangerous place, that’s not at all what you want! The last thing you need is for your images to be attributable to you! What you need is proof your images are real without surrendering your anonymity. That’s what Apple’s Reference Image does!

If you’re a content creator and you’re worried about being accused of faking or stealing your content, you want both! You probably won’t get C2PA right down to the camera level, but you can very easily get it to attribute all your hard work editing the raw material into your final product to you! And, if you shoot on an iPhone, you can use Apple Reference Image to prove the inputs to your work were real rather than AI-generated.

And finally, if you’re just a regular social media poster fed up with being accused of faking every cool thing you manage to snap and share, you can use Apple Reference Image to shut the trolls up! Or, if you’re unfortunate enough to need to capture some evidence for an insurance claim or a court case, you can use Apple Reference Image to prove you didn’t fake your evidence!

Final Thoughts

For most people, Apple Reference Image is all they need, and it allows us to prove our images aren’t faked without sacrificing any of our privacy. For creators and journalists, they can use C2PA to verify their claims to their work.

Apple hasn’t reinvented the C2PA wheel; it’s solved a different problem for different people in a different way. How very Apple!

Epilogue — Reference Image Experiments

After we finished recording our conversation, myself and Allison did some experimentation, and we discovered some interesting practical details.

Firstly, because I Reference Image is not available in Europe at the moment (October 2026), the first thing we wanted to try was verifying a Reference Image on another Apple device. Initially, it looked like this was not working, at least not for me in Europe, but that’s thankfully not the case. The reason it didn’t work initially is that by default, the Photos app doesn’t embed the Reference Image when exporting photos. This meant that when I first imported Allison’s example photos, I couldn’t see the Reference Image at all. The next day I needed to check the Photos app settings for something un-related, and noticed a setting for controlling whether or not to embed Reference Images when exporting, and it was off by default. Once Allison enabled that setting her exported examples worked as we had hoped — even in Europe the images were marked as containing an embedded Reference Image, and I could toggle between the edited version and that Reference Image.

My experience appeared to be identical to Allison’s — When I tapped on the Reference Image badge I got the I dream of Gene effect Allison described in the conversation, and then I could tap the image to toggle between the edited version and the Reference Image. To really show the feature in action Allison used Apple’s AI features to remove a few objects from the image, and as I toggled the objects appeared and disappeared, clearly showing the image has been altered.

Allison was also curious whether simply importing an image with an embedded Reference Image would be enough to trigger the creation of the Reference smart album — it is! As soon as I imported Allison’s exported image the new category appeared under the Types dropdown along with Selfies, Live Photos, Portraits, and more.

Finally, we verified that Private Cloud Compute is needed to generate and verify Reference Images — turning on Airplane Mode disables the feature!