Hash collision in Apple NeuralHash model (github.com)

Article summary

A hash collision has been found in Apple's NeuralHash model, which is used to detect child sexual abuse material (CSAM). The collision was demonstrated with two images that produce the same hash value. This raises concerns about the potential for false positives and the ability to manipulate the system. The issue is being discussed in the context of Apple's plans to scan iCloud photos for CSAM.

Main themes

• CSAM detection
• Hash collisions
• NeuralHash
• False positives
• Misuse and abuse
• Privacy and security
• Social and political implications

What commenters say

• The existence of a hash collision in Apple's NeuralHash model poses a significant risk of false positives and potential misuse.
• The system's design and the use of a blinded hash table make it difficult to exploit the collision for malicious purposes.
• The ability to generate collisions could be used to frame individuals for possessing CSAM, which could have severe consequences.
• The fact that the hash collision can be demonstrated with innocuous images, such as pictures of dogs, suggests that the system may not be effective in distinguishing between legitimate and malicious content.
• The use of NeuralHash for CSAM detection is a flawed approach that could lead to unintended consequences, including false accusations and reputational damage.
• The potential for abuse of the system by powerful individuals or governments is a significant concern, particularly in the context of political repression or censorship.
• The discussion around NeuralHash highlights the need for a more nuanced and transparent approach to CSAM detection and prevention.
• The risks associated with NeuralHash are not limited to the technical aspects of the system, but also involve broader social and political implications.