Code in the fifth beta of iOS 27 points to a system Apple appears to be developing under the name Apple Reference Image, which would write provenance data into a photograph at the moment the shutter fires. The purpose is easy enough to state: give someone a way to demonstrate that a picture came out of an iPhone camera rather than out of a generative model.
Nothing about it is switched on yet. A privacy disclosure in the beta says the feature will ship disabled, and that only photographs taken through a new Reference option inside the Camera app would carry the information needed to authenticate them later. The route to enable it, according to the code, runs through Settings, then Camera, then Reference Image, then Reference Mode.
Verification would not happen quietly in the background either. Tapping a Reference badge on a photo sends the untouched image together with its embedded data — sensor signatures, the window in which the shot was captured, hardware identifiers unique to the device — to Apple's Private Cloud Compute servers, which hand back an authenticated copy carrying an identifier of its own.
Apple would not see the photograph itself during that exchange. It could, however, receive enough sensor information to refuse authentication for images traced to a sensor known to be compromised, or to withdraw authentication it had already granted to earlier pictures from that sensor. Once authenticated, an image can be viewed as such on an iPhone, an iPad or a Mac.
The underlying idea is not new, and Apple is late to it. C2PA Content Credentials already do much the same job, recording how and where an image was produced and whether generative tools touched it along the way. Canon, Nikon, Sony, FujiFilm and Leica have been folding support into their hardware, and the cameras on Google's Pixel 10 handle the standard as well. Apple has conspicuously stayed out.
The reasonable inference is that Apple thinks it can do better on its own. C2PA has not built a reputation for reliability, and a company that controls the sensor, the operating system and the verification infrastructure end to end has a genuinely stronger hand than one implementing an open standard across hardware it does not own.
There is a second reading of the move worth noting. Building provenance into the camera is a bet on labelling work made by people rather than chasing work made by machines after the fact. That is the harder problem inverted, and it is roughly the position Instagram's Adam Mosseri has taken — that marking human-made content is likely to prove more tractable than reliably detecting everything a model produced.
