2 ms·
Right. But that means, in my eyes, that Signal's official builds are _exactly as untrustworthy_ as Signal's server is; if I couldn't trust Signal if they said t
by mort96 4y ago
Right. But that means, in my eyes, that Signal's official builds are _exactly as untrustworthy_ as Signal's server is; if I couldn't trust Signal if they said they didn't log non-E2EE messages, I don't see why I can trust Signal when they say they don't have a way to retrieve plaintext messages from their client build. If we're talking about liability if claims are false, I should be fine using any chat platform which claims to not log messages, right?
The RoTT thing isn't that relevant IMO; if we're under an RoTT-style attack, everything is off. But crucially, that's a different threat model; I can trust my compiler independently from Signal. I don't feel like I can trust Signal's client independently from Signal's server; either Signal as a whole is trustworthy, or they're not.
Now I agree that if we had some kind of verifiable or reproducible builds, E2EE would make sense. If anyone could download Signal's source code, build it, and verify that the binary is identical, we would have the "normal" FOSS defenses, where we could assume that some security researcher would do that now and then.
- woodruffw 4y ago> If we're talking about liability if claims are false, I should be fine using any chat platform which claims to not log messages, right? It's a fine point, but I'd argue that Signal's promises are still "worth more" than a competitor who merely promises to not log without any technical backing: the latter is a purely social promise, while the latter is a social promise rooted in a cryptographic claim that, with the right skills, you could verify. The RoTT threat model (which I agree is extreme!) is really not that far off: when you download the official Signal app, you're trusting, modulo the integrity of the app store, that Signal has either backdoored everyone or nobody. In other words: they would need to deliver the weakness to all users, increasing the risk of discovery, or to none of them. Binary transparency makes that gamble riskier still, by forcing Signal (or the app store) to publish signatures for each build before clients accept them for installation. Computers will never perfectly eliminate the latent social trust required when running someone else's code, which we all do even when we run any code at all (even our own). But Signal's current model is about as good as it gets for laypeople: you can verify the source, and the parts you can't verify require significant malicious investment and/or exposure risk. For the more (potentially rightfully?) paranoid among us, building from source and comparing the binaries is about as good as it gets for now.
- mort96 4y ago> the latter is a social promise rooted in a cryptographic claim that, with the right skills, you could verify. How would one do that? Do you mean if you literally decompiled the binary and read through the machine code and built up a solid understanding of what everything does (and did this after every app update)? Or is there something more practical you're referring to? > when you download the official Signal app, you're trusting, modulo the integrity of the app store, that Signal has either backdoored everyone or nobody The way I imagine this attack working, Signal would have code in everyone's app which lets the server ask for DMs, but the server would only ask for DMs when there's a reason to do so (such as when a government requests it). That way, yes, everyone is backdoored, but there's no suspicious network traffic or anything which would give it away. I'd say Signal can be pretty confident that nobody will read through the machine code and make a solid enough mental model of how the entire system works to notice such a backdoor.
- woodruffw 4y ago> Or is there something more practical you're referring to? I would probably compile the reference source and use BinDiff[1] (or a handful of other tools) to compare it with the official binary. I've used BinDiff successfully for similar purposes in the past; it's also a popular tool for finding N-days after patches go out. Signal could patch in that kind of backdoor, sure. But it's still globally disclosed in that case: you're revealing your backdoor to every prospective reverse engineer (there's a lot of them!), even if you only selectively enable it. Signal can probably be about as confident that nobody is reading their machine code as Chrome can, which is to say not confident. And note: you don't need a solid enough mental model of how the system works at the machine level. You only need a differential understanding of what the public code says and what the machine code does, which is much easier. Is it possible? Sure. But it's much, much higher risk and visibility than every other technique available to nation-state adversaries. [1]: https://www.zynamics.com/bindiff.html https://www.zynamics.com/bindiff.html