[skeezyboy]

1. Of course its possible, youd just get back encrypted data. This doesnt make it impossible

2. You mean "Reliable classification of client-side scanned images is impossible", although you dont actually define reliable. This is besides the point, Im not talking about the actually feasibility of this on a political level, Im asserting a specific technical point that client-side scanning is 100% possible for e2e apps

[ivan_gammel]

That specific technical point is trivial and is not worth discussing it. Of course you can „scan“ a data stream, but what’s the point if it doesn’t yield any meaningful results?

The only acceptable scanning process here is the one that produces only true positives, no collateral damage. This is what I call reliable.

[skeezyboy]

> The only acceptable scanning process here is the one that produces only true positives, no collateral damage. This is what I call reliable.

well then reliability is impossible, you must accept errors

[ivan_gammel]

> well then reliability is impossible, you must accept errors

Nobody should accept errors. Client-side scanning simply must not happen. It’s mathematically dumb idea.

[fluoridation]

1. If the client application is hashing ciphertext, its hash will not match any known offending hashes, even if the plaintext is a known file.

I don't understand why someone would go through the trouble of using WhatApp to pass around separately-encrypted files instead of using anything else, though.

2. It's also "technically possible" to do the scanning server-side, on the encrypted stream, and flag anything that by chance matches a known hash.