5 ms·

Could have been written yesterday; it seems pretty fresh to me. I regret TFA's counfounding of the role of CAs and the role of DNS. The bug with CAs is that th

Could have been written yesterday; it seems pretty fresh to me. I regret TFA's counfounding of the role of CAs and the role of DNS. The bug with CAs is that the role exists at all; the bug with DNS is down to implementation. CAs are a solution to the problem of how to securely introduce two parties that don't know one another. That's a fundamental problem, and CAs aren't the only solution, but there aren't many. IMO, CAs are a really awful solution. As TFA says, they are unreliable and expensive to operate. WoT is a better solution, but expensive to set up, which is perhaps why it's unsuccessful. [Edit] WoT perhaps isn't a better solution; you still have to be able to find the trusted certificate, and ultimately that means you still need a trusted introducer. That means a TTP; or you meet in person, which is in conflict with the demand that this is someone you haven't met. But so what if you've met them? It comes down to bank statements and driving licences, which amounts to official government-issued ID. But suppose what I want to do or say is unrelated to my official ID, and I'd prefer to certify idependent of government and officialdom? Suppose I'm stateless, and have no government? I'd like self-signed certificates to be treated by browsers as first-class citizens. [Edit 2] Given certificates done right, everything else falls away; you can do what DNS does using your certificate. You can sign a DNS zone, or sign any other statement you want, and the certificate itself can direct people to those signed statements. DNS solves another problem: how to map names to [data], where [data] is usually an address of some kind. It assumes a particular naming system, and that system of names isn't fundamental. It's a layer on top of lower-level addressing systems, and could be replaced without affecting lower layers. It's dirt-cheap to operate; what's expensive and buggy is the market for names.

JohnFen 3y ago
> IMO, CAs are a really awful solution. I agree. But the only solutions that can work are ones everyone agrees to use and, for right now anyway, everyone's settled on that model. There isn't much choice.
- rainsford 3y ago
  I disagree the problem is a lack of choice, because what realistic alternative choices actually exist? Web of trust a-la PGP key signing parties is a ridiculous solution for people who aren't super nerds (and even for super nerds it's pretty bad). Trust on first use works OK-ish in SSH, but it's hard to see that as a realistic solution for trust as a whole if you try to scale it up to every user needing to access arbitrary servers. Maybe CAs are a bad solution, but if so it's in the sense of the quote about democracy being the worst form of government except for all the other ones. At the end of the day, the CA system has managed to deliver working cryptographic trust at a scale that galatically surpasses anything else.
  - JohnFen 3y ago
    > what realistic alternative choices actually exist? Yeah, that's a problem. I certainly can't think of one. But even if there were, it would still be an enormous struggle just to shift to using it at this point. That struggle also means that there isn't a huge amount of effort being put into finding a better way. The CA system certainly has great benefits in terms of delivering reasonable security in a relatively convenient way, but that doesn't take away from the fact that it still sucks in ton of different ways that are inherent to the idea. It also can't be used (at least not without great struggle) in some situations.
    - rainsford 3y ago
      I definitely agree that there is an inertia problem, as there is with a lot of technology. But I also think changes can happen quicker than people think if new alternatives are introduced that solve fundamental problems with the entrenched solution. Arguably this actually has happened with the legacy CA system, people just don't think of it that way because it looks like Let's Encrypt and certificate transparency instead of something involving blockchain or whatever. LE and CT were huge disruptions in the CA space and in my view fixed a lot of the problems with the CA system, even if they didn't totally get rid of it. To borrow your excellent phrasing, they made it suck in a lot fewer ways than it used to. Maybe some complete replacement could do better, but there is a lot to be said for iterative improvement of an existing solution instead of hoping for a fantasy total replacement.
      - JohnFen 3y ago
        Interesting that you bring up LE as an improvement of the CA system. I view it as a fairly messy hack (with its own problems) around some of the faults of the CA system. I almost brought it up as an example of some of the faults of the system. We both see the same thing, but through different lenses. I have no point here, I just found that interesting.
  - denton-scratch 3y ago
    Yeah, I didn't mention TOFU because it's a non-starter if you're trying to get a secure introduction to someone you don't know. The fundamental problem is that "identity" is problematic. How do you know someone is who they say they are? Perhaps you don't really know who you are. That's why I want a space for a certified identity that isn't tied to a government ID. Like, an instance of some software might claim an ID. Arguably, a domain is an ID. I don't see what's wrong with something or somone having multiple IDs that aren't linkable.

JohnFen 3y ago

> IMO, CAs are a really awful solution. I agree. But the only solutions that can work are ones everyone agrees to use and, for right now anyway, everyone's settled on that model. There isn't much choice.

rainsford 3y ago
I disagree the problem is a lack of choice, because what realistic alternative choices actually exist? Web of trust a-la PGP key signing parties is a ridiculous solution for people who aren't super nerds (and even for super nerds it's pretty bad). Trust on first use works OK-ish in SSH, but it's hard to see that as a realistic solution for trust as a whole if you try to scale it up to every user needing to access arbitrary servers. Maybe CAs are a bad solution, but if so it's in the sense of the quote about democracy being the worst form of government except for all the other ones. At the end of the day, the CA system has managed to deliver working cryptographic trust at a scale that galatically surpasses anything else.
- JohnFen 3y ago
  > what realistic alternative choices actually exist? Yeah, that's a problem. I certainly can't think of one. But even if there were, it would still be an enormous struggle just to shift to using it at this point. That struggle also means that there isn't a huge amount of effort being put into finding a better way. The CA system certainly has great benefits in terms of delivering reasonable security in a relatively convenient way, but that doesn't take away from the fact that it still sucks in ton of different ways that are inherent to the idea. It also can't be used (at least not without great struggle) in some situations.
  - rainsford 3y ago
    I definitely agree that there is an inertia problem, as there is with a lot of technology. But I also think changes can happen quicker than people think if new alternatives are introduced that solve fundamental problems with the entrenched solution. Arguably this actually has happened with the legacy CA system, people just don't think of it that way because it looks like Let's Encrypt and certificate transparency instead of something involving blockchain or whatever. LE and CT were huge disruptions in the CA space and in my view fixed a lot of the problems with the CA system, even if they didn't totally get rid of it. To borrow your excellent phrasing, they made it suck in a lot fewer ways than it used to. Maybe some complete replacement could do better, but there is a lot to be said for iterative improvement of an existing solution instead of hoping for a fantasy total replacement.
    - JohnFen 3y ago
      Interesting that you bring up LE as an improvement of the CA system. I view it as a fairly messy hack (with its own problems) around some of the faults of the CA system. I almost brought it up as an example of some of the faults of the system. We both see the same thing, but through different lenses. I have no point here, I just found that interesting.
- denton-scratch 3y ago
  Yeah, I didn't mention TOFU because it's a non-starter if you're trying to get a secure introduction to someone you don't know. The fundamental problem is that "identity" is problematic. How do you know someone is who they say they are? Perhaps you don't really know who you are. That's why I want a space for a certified identity that isn't tied to a government ID. Like, an instance of some software might claim an ID. Arguably, a domain is an ID. I don't see what's wrong with something or somone having multiple IDs that aren't linkable.