- Published on
Apple Reference Image, but open source and using steganography
- Authors
- Name
- DeepHorus
Apple has introduced Apple Reference Image, a system that cryptographically proves a photo was actually taken by a camera rather than generated by AI. The September 2026 announcement responds to the growing difficulty of distinguishing real imagery from synthetic content. Independent developers have already built a similar open-source alternative using steganography.
The project came out of the Recurse Center, where María Benavente and Alex Hornstein created a proof-of-capture camera. Their device uses a Raspberry Pi Zero, a display board, an ATECC608 crypto chip, a shutter button, and a 3D-printed enclosure. It costs under $100 and is fully open source.
The approach avoids the losing race of AI detection. Benavente explains that every detector improvement becomes training signal for the next generator. Instead, the camera proves authenticity at capture time by embedding a signed perceptual hash into image pixels. This steganographic watermark survives compression and resizing, unlike metadata signatures.
The ATECC608 chip signs the hash while keeping the private key inaccessible, even to the owner. Tampering causes the chip to lock itself. The current version spreads the watermark across the image using DWT and DCT, surviving WhatsApp-grade compression.
Apple's system also signs pixels at capture time, but stores the signature in a "digital negative" through Private Cloud Compute. Apple has not yet made its verification flow public. Benavente criticizes Apple for ignoring the existing C2PA standard used by Nikon, Sony, Leica, and Adobe, and notes that the root of trust remains inside Apple's cloud.
No solution fully solves the problem. A screen attack can still produce a signed photo of a fake by photographing a display. Still, Benavente welcomes Apple's interest in authenticity. The open-source project lets anyone build a camera and inspect the hidden signature.
