40 ms·
A few thoughts about Signal’s Secure Value Recovery
- tptacek 6y agoI don't like SGX at all and wish Signal wasn't using it (they presumably could do something more like what Apple did with their HSM cluster system, which had exactly the same purpose of defeating brute-force attacks on low-entropy secrets). But this is a weird article. The "Addendum" moots much of what precedes it, and both the main body of the article and the Addendum postdate the debate Green and Moxie had on Twitter. The whole thing appears to boil down to: * Yes, every one of Signal's mainstream competitors store in serverside plaintext the information Signal protects with SGX, meaning that none of the microarchitectural attacks Green discusses could reduce the security of Signal below that of realistic alternatives. * And, you can now disable PINs on Signal altogether (even before doing that, you could effectively disable them by setting and forgetting an intractable passcode, but I agree that UX would have been inadequate). * So what we're left with is the claim that Signal could in the future enroll more data in SGX beyond contact lists. Which of course, they can, and always could have. As the article hints early on, other messaging platforms have no qualms storing message data in places that provide something much less than end-to-end-encrypted security guarantees; Signal could for the last several years have been implementing those same user-pleasing features without SGX without risking security parity with other messengers. They didn't. Why am I meant to believe they're going to do it now? You're in a really weird place when you're arguing that Signal, the messenger that requires people to use phone numbers as identifiers (that this infuriates every nerd on the Internet is not lost on Moxie and Trevor), that has an encrypted tunnel set up just to share GIFs to foil traffic analysis and wouldn't roll that feature out without it, that only just in the last 2 years got user profiles because of the privacy concerns --- when your core argument is that this team has somehow indicated bad faith about user privacy. I mean, it's just the opposite. (I'm no more clued in than anyone else on this stuff but my take on Signal's excitement about contact lists is that it's a step towards providing something other than phone number identifiers, which isn't just a big deal on the order of message backups, but maybe the biggest possible deal with respect to the UX of the service. But I have nerd bias on this, because I hate using phone numbers. It's possible it's less of a big deal than I think, and that ordinary non-nerd uses DGAF about phone number identifiers, since it's how they all message already.)
- psanford 6y agoOh thank you Moxie for finally letting us turn off the PIN nagging UI!
- Gaelan 6y agoMy concern, personally, is that Signal seems to be designing features with the assumption that SGX is sufficiently secure to be used to provide Signal's privacy guarantees. While yes, SGX is better than the nothing, "Intel has the keys and probably won't publish them" is very different from "only your device has the keys." To me, much of the appeal of Signal is that they actually back up their privacy guarantees with crypto, not just promise they won't read it. Given that (I assume) Intel has the remote attestation keys, SGX doesn't seem much better than any other promise-based privacy scheme. Additionally, I am generally very careful about sharing my contacts list only with apps I know won't upload it[1]. To me, silently starting to upload my contacts—even if they're protected by Actual Strong Crypto, but especially if it's a scheme like SGX that relies on humans doing the right thing—feels like a huge betrayal of trust. I'll be honest: I've never actually used Signal (I have the app installed, but nobody I want to talk to does). But I used to be enthusiastic about finding a chance to use it, but now it's more of a "probably better than any alternative" situation. [1]: It's probably silly, given that my Instagram and WhatsApp usage probably provides a better picture of my social graph than my contact list does, but oh well.
- matthewdgreen 6y agoSignal spent months and years designing a system for backups that had no opt-in or opt-out feature. They deployed it over a period of months, ignoring a lot of vocal criticism from people who said "this is fine, but why aren't you letting me opt out?" Finally, at the 11th hour, they hacked in an opt-out feature. Note that the opt-out feature doesn't even stop data from going to Signal's servers -- preventing this upload appears to be technically infeasible! Rather, the hack is to send the data anyway, but encrypt it under a random key. It does not bode well for Signal's security engineering and design philosophy that "don't upload data to a server, even when users understand and accept the consequences of that" is not something Signal even considered supporting, and their design is completely incompatible with the idea. If you want to think of this article as a post-mortem on a thing that went wrong at Signal, that's fine with me. Claiming that it shouldn't be written is just silly.
- upofadown 6y agoThe SGX enclave thing doesn't have to be really all that secure to solve a problem for signal. All it needs to do is be secure enough so that they can reject things like subpoenas. It changes the required level of cooperation from passive to active.
- lunixbochs 6y agoAndroid [1] and iOS [2] both have APIs and UI to associate a custom instant messenger protocol entry, e.g. "Signal": "signalusername" with a real contact. SVR is thus not required to store Signal handles in my phone native contact list, so why do Moxie and others keep saying this is somehow necessary? [1] https://developer.android.com/reference/android/provider/ContactsContract.CommonDataKinds.Im https://developer.android.com/reference/android/provider/Con... [2] https://developer.apple.com/documentation/contacts/cncontactinstantmessageaddresseskey https://developer.apple.com/documentation/contacts/cncontact...
- tptacek 6y agoPeople expect a consistent "buddy list" of Signal contacts on all the platforms they use, and across reinstalls.
- ThePowerOfFuet 6y agoWhy wouldn't contact syncing also solve this? It's just another CardDAV field, no?
- anigbrowl 6y agoSome people do. While this makes makes things much more convenient for cross-platform use or the case where someone loses their phone and doesn't want to lose access to all their signal data, it lowers security (conceivably) and trust (definitely) when they're forced to store the contact information through the requirement of a PIN.
- tptacek 6y agoThough the article would (I guess unintentionally?) lead you to believe otherwise, you are not in fact forced to store your contact information with Signal. The word "some" is doing an interesting amount of work in your comment, incidentally, since every other mainstream messenger just stores this stuff in something like a serverside SQL database.
- 6y ago
- Arnt 6y agoI'm quite surprised by the number of people who misunderstand this SGX stuff. Regarding SGX, Signal is its own attacker. This is a special situation. The SVR stuff has attackers; I could attack it to get your contacts, but my attack would not touch SGX. I could try a stream of password guesses, no SGX. SGX is only relevant if the attacker is Signal itself, presumably ordered by a court. AFAICT the actual security properties are therefore important, what matters is SGX' legal properties. SGX servers to let Signal say in court that "this data is stored in such a way that we can't read it and therefore we can't produce it in court; just ask Intel". If there is a secret way to break SGX and the court learns it, then a court could reveal that to Signal. "Here Moxie, take this super-duper vulnerability and use it." Eh. Or the court could order Intel to make a microcode update that disables SGX. Also eh.
- matthewdgreen 6y agoI don't know that you really need to put emphasis on "super secret way to break SGX". Any single attestation private key, combined with cooperation from Signal (or a compromised Signal operations admin) will suffice. SGX attestation has now been publicly broken twice in two years, first with Foreshadow and then with SGAxe. To give some color to what "publicly broken" means: there's currently a live Twitter bot put up by UM that signs valid attestations using an extracted (non-revoked last I checked) SGX attestation signing key [1]. The previous exploit, Foreshadow, also had one [2]. Enclaves with the very latest microcode updates will reject those attestations (and Moxie says Signal has installed those updates, which I believe), but this vulnerability is now several months old and has been in non-public disclosure for most of that time, long before microcode updates reached end-customers like Signal. This means that any of a team of grad students and professors who worked on this paper (and this is only one such team doing this work) had access to the ability to "super secretly break" SGX during that time. Based on the team's track record: I would personally bet $500 that some team regains this capability sometime during the next 12 months, using an as-yet-undisclosed new exploit. So we're not talking about "NSA levels of exploitation" here. We're talking about "can I pay a smart grad student" levels of exploitation. And one such exploit, combined with Signal's cooperation, means the loss of all backup data in an irrevocable and untraceable way. [1] https://twitter.com/SGAxe_AaaS https://twitter.com/SGAxe_AaaS [2] https://twitter.com/foreshadowaaas https://twitter.com/foreshadowaaas