gickhocq

I think I see your argument, are you saying that because the scanning involves transforming the content into a hash and comparing to a set of hashes, the problem is that this is resistant to third-party analysis? Not because of the overall behaviour of the scanner or indeed any other software on the device, which can be reverse-engineered using standard techniques, but because a database of hashes is inherently opaque as to the content it will match with?

I don't really see the problem with this. Scanning happens before the message is encrypted or after the message is decrypted (thus preserving the end-to-end encryption itself from compromise), occurs only on the device itself, and will report back only if known illegal content is detected.

It seems to be the best possible method of implementing such scanning. Shouldn't we be supporting this as a proportionate safety measure?