Apple's Client-Side Photo Scanning Algorithm Reverse-Engineered
Security researchers successfully reverse-engineer Apple's NeuralHash algorithm, discovering it already exists in iOS 14.3 and exposing critical vulnerabilities that allow for hash collisions.
Security researchers successfully reverse-engineer Apple's NeuralHash algorithm, which the company plans to use for client-side scanning on the iPhone. They discover that this code is already present within iOS 14.3, meaning the underlying technology is currently sitting dormant on millions of devices.
Initial testing reveals that while the algorithm tolerates image resizing and compression, it fails to handle basic alterations like cropping or rotations. Furthermore, researchers quickly produce the first hash collision by finding two completely different images that generate the exact same NeuralHash value.
This breakthrough paves the way for adversaries to generate innocuous images that the system incorrectly classifies as prohibited content. Critics point out that Apple fails to consider the adversarial context of the entire system, proving this scanning approach is a fundamentally flawed idea from the start.