Hacker News Reader

Comment by johansch | Hacker News Reader

You are also missing the point. The law mostly cares about intent and the prosecutor (or whatever it is called where you live) being able to convince the jury or judge of what they thought happened.

It's not like the sort of logical reasoning that happens when software engineers battle for the Right way to do something. :)

View on HN

No, I still maintain it can be engineered around. If enough people start random streams between themselves, is the government going to prevent that? Noise actually can be put into anything, it could be images, sound, random.org, whatever.

You can then embed crypto in those streams. When the prosecutors can't even distinguish intent then there's a problem. No third party may ever know someone exchanged encrypted messages.

So again, I'm not saying the law is fundamentally inconsistent or anything, it's just that it's pretty easy to make it unenforceable and undetectable.

> Noise actually can be put into anything, it could be images, sound

You just massively increased bandwidth cost and download times.

It's about time a Comcast rep showed up here. :-)

Cryptograms can't be compressed either (so the additional bandwidth requirement can be close to 0).

No, you're saying that people should add noise to everything, which steganographically masks the times when they add encrypted data instead.

The noise, which is added to everything, is not compressible so now everything takes a lot more badwidth.

See my comment on Generative Models above. In general there are too many degrees of freedom with modern communication channels for concealing data to be an insurmountable challenge, even if you don't have powerful Generative Models.

Keep in mind the law is targeting malicious actors, people that would likely try to use this sort of circumvention even if it isn't mainstream.

Which government?

In a shit sack world, the inquisition just points out that there is noise in a lot of your messages and you're done, you don't get a chance to wax philosophical about your intent.

In an even worse world, they just insinuate that you might be communicating secretly.

What is noise anyway? I can program a generative model and inverse generative model couple and then you can transmit data as a image that doesn't look "noisy" but has incompressible information (what I referred to as noise), which could be a cryptogram. The model might generate anything, a innocuous conversation, a picture of a house, whatever. For good models the "inquisition" cannot distinguish you from your neighbor in terms of suspicious activity.

This is just RubberHose, with the same flaws. (Increased bandwidth, and "they" don't care if it's noise or cipher text, they'll happily beat you until you give them what they want.)

https://web.archive.org/web/20121029045140/http://marutukku....

https://en.wikipedia.org/wiki/Rubberhose_%28file_system%29

https://en.wikipedia.org/wiki/Rubber-hose_cryptanalysis

No, you're misunderstanding my proposal (w.r.t. Generative Models). This is different from deniable encryption (where an eavesdropper will detect you may be encrypting something), this is traffic literally indistinguishable from normal traffic.

It only fails if the government knows a priori you are a criminal and inspects your hardware.

I thought about making a few proof of concept models for this, but I don't think there's real demand for it (and I'm lazy) -- steganography is probably sufficient for what most people use.

Reply on news.ycombinator.com