Cryptographer Matthew Green calls Apple's "trusted camera" photo signing scheme too fragile

On September 16, cryptographer Matthew Green posted a series of critiques of Apple's new Reference Image (ARI) "trusted camera" photo provenance mechanism, arguing that the signing-based content-credential scheme (similar to C2PA) is too fragile to authenticate high-value photographs. He acknowledged that deploying at the system level is a good direction, but said the security model does not scale and that it is only a matter of time before someone forges a "verified" photo that passes validation.

Confirmed

• Green's core objection is fragility: if even a small number of the millions of cameras are compromised, forgers can mass-produce fake photos with trusted signatures, collapsing the system's trust foundation.
• He argued Apple's description of how the system works is vague — likely only Apple itself can trace a photo back to a compromised camera, while the public cannot verify provenance on its own, limiting usefulness while adding new risks (e.g., an "alien photo with an ARI signature" the public cannot judge).
• He noted the scheme involves identity binding: tying camera identity to real-world identity can prove a photo came from the user's own camera rather than a third-party forgery, but strong default privacy is in tension with the security of such a large-scale AI deployment.
• He endorsed the privacy intent but said it does not make up for the verification gap.
• He offered a use case he does endorse: low-value economic transactions, such as insurers accepting photos after a car accident without sending an adjuster — a user could theoretically compromise their camera to fake photos, but the payoff doesn't justify the cost, so binding camera identity to real ID can cheaply deter such fraud.

Why it matters

ARI represents the industry's "trusted signature" approach to fighting AI-generated imagery, and Green's critique targets its fundamental assumption: a signature proves a photo came from a camera, not that the camera wasn't compromised. In high-value scenarios such as news evidence this gap could be fatal, while in low-value scenarios the cost-benefit calculus is what makes signatures meaningful — an important reference point for assessing the applicability of similar C2PA-style schemes.

2026-09-16 ~ 2026-09-16 · 7 related posts

• [source] Cryptographer Matthew Green: "trustworthy cameras" are too fragile to rely on — matthew_d_green · 2026-09-16
• [source] Cryptographer Matthew Green questions Apple's photo provenance signing for public trust — matthew_d_green · 2026-09-16
• Matthew Green: photo provenance signing fits low-value transactions like insurance claims — matthew_d_green · 2026-09-16
• Cryptographer Matthew Green: camera identity tied to real ID can curb insurance fraud cheaply — matthew_d_green · 2026-09-16
• Cryptographer Matthew Green: Strong Default Privacy Clashes With Large-Scale AI Deployment Security — matthew_d_green · 2026-09-16

2 near-duplicate retellings: matthew_d_green · matthew_d_green