12 Factors of Threat Modeling
With Chris Romeo, Izar Tarandach, and Matt Coles
Chris, Matt and Izar share their thoughts on an article published by Carnegie Mellon University’s Software Engineering Institute. The list from the article covers various threat modeling methodologies such as STRIDE, PASTA, LinDoN, and OCTAVE methodology for risk management.
About this episode
Chris, Matt and Izar share their thoughts on an article published by Carnegie Mellon University’s Software Engineering Institute. The list from the article covers various threat modeling methodologies such as STRIDE, PASTA, LinDoN, and OCTAVE methodology for risk management. They emphasize the importance of critical thinking in the field, provide insights into strengths, applications, and limitations of each method, and highlight the significance of annotated threat models for application security.
Mentioned in this Episode:
Article: https://insights.sei.cmu.edu/blog/threat-modeling-12-available-methods/
Podcast episode: Nobody's Going to Mess with Our STRIDE https://www.youtube.com/watch?v=TDFRe_icFmY&pp=ygUSdGhlIHNlY3VyaXR5IHRhYmxl
Follow AI Security Table:
➜ Home: https://securitytable.ai/
➜ X: https://x.com/AISecurityTable
➜ LinkedIn: https://www.linkedin.com/company/ai-security-table/
➜ YouTube: https://www.youtube.com/@AISecurityTable
Transcript · 46 min conversation
0:09Chris Romeo The audience is wondering, why did we enter the Security Table Podcast to a laugh?
0:18Matt Coles We—
0:18Chris Romeo and that is perhaps because we declared that there's a 2-drink minimum going forward.
0:23Izar Tarandach It's part of our podcast.
0:25Chris Romeo Which, yeah, there's one. It's early in the day. So actually, oh, there's a two-drink. Oh, excellent. Look at this. Matt is meeting the two-drink minimum, which is excellent. So he'll be next on stage here for Application Security Standup is what we call it. It's not the standup that you imagined from Agile. It's standup comedy from AppSec people. Ready? Here we go. Okay. We would not make people listen to that. That would be terrible. Very bad.
0:55Izar Tarandach Very bad.
0:55Chris Romeo And unbearable. A standup segment by a bunch of AppSec people, standup comedy.
1:01Izar Tarandach How would that begin? Uh, so threat modeling, right? What's the best—
1:06Chris Romeo that's such a—
1:09Izar Tarandach what about threat modeling?
1:15Chris Romeo And then there may be some examples. Yeah.
1:20Matt Coles I just flew in from ThreatModCon and Boy, am I already tired.
1:24Izar Tarandach Thank you.
1:27Chris Romeo Wow. That is comedy gold, ladies and gentlemen. Comedy gold. You, I mean, where else do you get comedy like this? I mean, this is, I would say we're here until Thursday, but today's Friday, so yeah, we're here all week.
1:44Matt Coles It's a buffer overflow.
1:48Chris Romeo It keeps getting worse. That's the problem with every joke.
1:51Izar Tarandach Very bad.
1:52Chris Romeo We fall, we fall further and further. Okay. Well, we weren't just here for the purposes of standup comedy, but that was how we began. We want to talk about an article that gets referenced a lot. And so we want to unpack what this article says and offer some commentary on it. It was written in 2018. It's called Threat Modeling: 12 Available Methods. And you'll find it on the Carnegie Mellon University Software Engineering Institute, SEI. It's on their blog. And it's, it's, it's caught our attention and, and we've all seen it at different times in the past and seen people referencing it. And there's some, there's some interesting things that are described as threat modeling in here. That's really what we want to unpack. So we'll just kind of scroll down through the list and, and deal with each one as they go. The first one is obviously Stride and associated derivations, which is the best possible answer. You could literally have ended the article right there. You could have just put conclusion after Stride and associated derivations. What do you guys think?
3:01Izar Tarandach Agree? I mean, we spoke a lot about Stride on the last episode, so everybody knows that, uh, we like it, we respect it, we endorse it. We suggest it sometimes.
3:15Matt Coles We use it many times, but, uh, it's sort of the underpinning for most other methods, right? It, it, it forms a, it forms a basis for a lot of what we do, right? We look at interactions or elements or both. And so stride is that, that core.
3:33Izar Tarandach We just, or, or, or at least the newest methods always measure themselves against Stride.
3:39Chris Romeo Yeah.
3:40Izar Tarandach When they get put out, right?
3:42Chris Romeo Because of its simplicity, right? Like it's got, it's such a simple core to work from that it's, it's, it is really the measuring stick of other approaches to threat modeling. But as Izar said, we, we really spoke about it a lot last episode. And so let's just let people go back and look at that previous episode for our love fest with Stride.
4:07Matt Coles And now before we, before we move forward, we might look at STRIDE as the base class for, if we look at a base class and a derived class for other threat modeling methodologies, there is STRIDE itself is actually a derived class from a methodology that is not mentioned here.
4:22Chris Romeo Hmm. Which one is that?
4:23Matt Coles Right. Well, the Four Question Framework, of course.
4:26Chris Romeo Well, STRIDE came long before the Four Question Framework.
4:29Matt Coles Before, yeah.
4:29Izar Tarandach But, but, but Matt has a point. Matt has a point. After STRIDE came in because of its simplicity, because of the methodology, because of the way that it was being used. Adam was able to, in the way that he does so well, subsume things to like this kernel of what actually needs to happen and what it is and come up with the Four Questions framework, which I would say, and I think that Matt just said, that everything else comes from the Four Questions that actually, by the way, of Stride. So I think that Matt is right.
4:58Chris Romeo Yeah. Yeah.
4:59Matt Coles Four Question Framework abstracts Stride. Stride then implements. The, the, the parts that are necessary and then additional models and frameworks build upon that. So in this hierarchy, Stride came first in, in chronological order, but the Four Question Framework effectively comes first in the framework.
5:17Izar Tarandach I hear dependencies. Do you have an S-bomb for that?
5:22Matt Coles I have, I, I, I have something choice, choice, some, I have some choice graphic for you going off screen at the moment.
5:31Izar Tarandach That's not the best.
5:33Chris Romeo Uh, that is exactly, you know, it'd be a super dashboard if you could see your DAS results and your SBOMs side by side correlated, a whole new category I just invented with, with a, with an AI, uh, looking at both and comparing.
5:47Izar Tarandach Yeah.
5:47Chris Romeo You don't even have to look at it. Wait, could I see?
5:50Izar Tarandach I'm fucking saying upgrade me.
5:52Chris Romeo Could you sell a product? Could you sell a product? That you claim was an AI agent looking at two different things.
6:01Izar Tarandach But yeah, that's everything. Please check the request. We'll just have to blend here.
6:05Chris Romeo People are like, well, how do I log into it? You don't.
6:08Izar Tarandach You don't.
6:10Chris Romeo AI just does it. And at the end of the day, you're like, you're like the ultimate grifter because there is no product behind it. It's literally just somebody buying something. You're like, oh no, no, it's the GenAI looking at all the things.
6:21Matt Coles You don't have to log into it.
6:23Chris Romeo Don't worry.
6:23Matt Coles I think Ezzar just described Cobalt Strike or something.
6:25Izar Tarandach So Chris, next week when you come back from RSA, tell us what the name of the product already is. Okay.
6:32Chris Romeo I'll be on the lookout to see if somebody's already invented it, but all right, we should keep going. We're only, we've only cleared the first hurdle of 12.
6:40Izar Tarandach The second one comes, STAAR, which I would say is the total opposite of STRIDE, right? So if we're talking about. A methodology that is too open, too generic, sometimes has problems in, in drawing boundaries. And I don't mean trust boundaries, but boundaries of the process and, and eliciting stuff and coming up with like a clean black and white format of things. I think that PASTA is completely on the other side. It's extremely detailed. It's based on statistics, probability, $1 billion amounts being counted. And it takes a long, long, long time to get done. And you need, in my opinion, you need a lot of expertise outside of realm of actually security and threat modeling. You actually need to do some heavy work, some heavy lifting to do Buster right. And I'm sure it has its uses.
7:44Matt Coles I really want to, I'm going to take a, probably a different, I want to take a different approach here. I think PASTA is a lot like Stride. If you look at steps 3, 4, and 5.
7:59Izar Tarandach Interesting.
8:00Matt Coles So I think, and I don't know for sure, I'm going to, I'm going to put something out there and probably get a lot of hate mail on this, but PASTA is not strictly a threat modeling approach.
8:11Izar Tarandach Yeah.
8:11Matt Coles Yeah, it's more than threat, just threat modeling, at least as we were talking about threat modeling today. If we look at the Four Questions Framework, we look at STRIDE, we build a model, we analyze that model, we look for security vulnerabilities and weaknesses and ways that attackers are going to come in. And that's part of PASTA, but there's more to it than that, right? There's requirement traceability and there's attack chains and attack modeling and there's risk and impact analysis and So, Posta uses a threat modeling approach in part.
8:42Izar Tarandach And more.
8:43Chris Romeo And more.
8:44Izar Tarandach Yeah.
8:44Matt Coles And that may be part of not only the confusion about its use, but also the complexity involved because it's trying to do more than what, what we would look at from a traditional threat modeling.
8:54Izar Tarandach Yeah.
8:54Chris Romeo But I look at it, I look at it though, and it's, it's complicated. There's a lot of steps to it and it just has a waterfall-y feel. To it.
9:05Izar Tarandach And, and, and that comes from the fact that you have to have very well-defined what is it that we are building before you even start so that you can decompose the whole thing into its little pieces.
9:15Chris Romeo Which is very Waterfall-y.
9:18Izar Tarandach And I think that even if you repeat steps 3, 4, and 5, which by the way, for people who are not familiar, 3 is application composition, 4 is threat analysis, 5 is vulnerability and weakness analysis. So if you close those 3 in a loop that happens during development and then you continue the rest at the end, I, I don't think that you end up with something that's immediately useful.
9:40Chris Romeo And you couldn't, for example, for a developer, you couldn't skip 6, ATT&CK modeling, ATT&CK service analysis, ATT&CK tree development, right?
9:47Matt Coles You couldn't, well, because that feeds your 3, 4, 5, and 6 would have to, that feeds your, that feeds your risk and impact analysis. But you could, in theory, do 3, 4, and 5. Sort of on a cycle, right? Building up that knowledge base, updating your attack models as you go. So there are probably ways to facilitate this for an iterative development approach, whether you can go full DevOps, probably not. But, uh, but this is really good. I mean, to benefit, for benefit for Pasta here, it is very structured. And very detailed, which means if you're in an environment where you need that level of, of, of, um, specificity, then it's probably a good approach. Uh, or at least you can benefit from that approach.
10:34Izar Tarandach And correct me if I'm wrong, but I think the PASTA was one of the first methodologies that included threat intelligence as an actual thing.
10:43Matt Coles Even back then. It's more than just—
10:45Izar Tarandach Yep.
10:46Matt Coles Right. So I mean, if we look at the risk management framework from, uh, from NIST, follows this approach very, very, very similarly, not necessarily the same, right? But it tries to do very, uh, very structured analysis. The traditional, I guess the more traditional threat modeling where you look at your requirements and you look at your threats and then you figure out what your security requirements are, right? And build and start building systems from that. So I can see the value in here. I don't know, I haven't used it myself, or at least I have not used it intentionally myself, but I think it has value for some.
11:29Izar Tarandach I agree with you that there is lots of value here for some, for some verticals, for some ways of working. But just to illustrate, a couple of months ago I gave a workshop. And, uh, I spoke about strides and, okay, let's try this thing. You have 20 minutes to try this thing. Then I spoke about pasta and I said, okay, let's try this thing. You have 20 weeks to try this thing.
11:56Matt Coles So it, it's, it's basically the, the, the, we poisoned the audience right there, there, is all right.
12:03Izar Tarandach I mean, no, but seriously, like, I, I mean, the vertical is why it's faster. Right. It was, it was hard to digest. The, the, the thing is that the, the things that make PASTA good for a certain vertical, in my opinion, make it not so good for all the other verticals.
12:26Chris Romeo Yeah.
12:26Izar Tarandach So today, if I come to a developer team who needs the actionable part of threat modeling to happen in the way that they develop things, and I put PASTA on the table, It's not going to end well.
12:40Chris Romeo Yeah, I'm the same. I'm, I'm coming out of the same way. I couldn't imagine taking PASTA to a development team and trying in an Agile where we're, we're working, we're using Agile for scheduling and we're using DevOps for pushing things as fast as we possibly can to production. I don't see how this factors into that world. I don't think, I just, it, it's too heavyweight. I need something lightweight. I need something quick, fast, get what they need to get done, identify some threats during the story planning process or during the initial design, make some changes to design, and then keep cranking, get stuff up into production.
13:19Matt Coles Yeah, I definitely, I definitely see that. If you're building, if you're building hardware or you're building, you know, medical devices or autonomous vehicles or space shuttles, this is probably a, a something that you can still benefit from.
13:32Chris Romeo Yeah, because you're still on a waterfall. Those, those things are still built on a waterfall approach to the, the way the project's set up, right? They start with requirements, they go to a design phase, they go to a build phase, they go to a testing phase. Sometimes they circle back and start the whole thing over, which is why those types of things take years to build versus we can put a feature into production this afternoon. Yeah. In the companies that we work with.
13:58Izar Tarandach It's not only that, but a lot of the, the initial data or the auxiliary data that PASTA requires to generate its, its results, most people don't have that. I mean, it's asking for probabilistic chances of a specific attack occurring.
14:16Matt Coles Well, so can, can I, can I just make a note that we'll get to that eventually when we talk about risk management practices?
14:23Chris Romeo Yes.
14:23Izar Tarandach Yes, yes, yes. And in risk management, that's fine. But if you ask for, for me, if you ask from most teams that as a, a get-go thing, like, let's start from here.
14:33Matt Coles Mm-hmm.
14:35Izar Tarandach It doesn't work that way. Okay. Now we have Kev, we have IPSS, all that good stuff, but they live in a different later realm. All right.
14:42Chris Romeo Let's keep moving. We're only 2 in now and we're Lindens.
14:47Izar Tarandach Linden.
14:47Chris Romeo We love Linden.
14:48Matt Coles We love Linden.
14:49Chris Romeo And it's not just because we're all friends with the wonderful Dr. Kim Wutz. Who was one of the primary people behind Linden, and I believe the smartest privacy threat modeling person on planet Earth.
15:01Izar Tarandach Hear, hear.
15:02Chris Romeo And so that's why, but Linden is a, I think you can see the inspiration of Stride in Linden, but it's completely focused on privacy. So linkability, identifiability, non-repudiation, detectability, disclosure of information, awareness, non-compliance. These are the categories of potential privacy challenges, and Linden provides that methodology. So, uh, I'm a giant fan. I think we're all giant fans of Linden. And just, I think we're all in the same boat. We're just trying to learn more about privacy. That's where I keep—
15:31Izar Tarandach well, then I'll drop a note here. Uh, recently, in ways that don't, don't come in scope, I was talking to someone about, uh, privacy. And yes, the, the— I, I got a look back and something like CIA, we have confidentiality. Why do I need privacy? And of course, I use Lindon and all that to explain the difference, but then I walk back to, hey, this could have been me a couple of years ago. Because there's a lot of people in security who have that mentality that confidentiality covers privacy and you don't have to deal about that. And especially talking to Kim, I think it, it's, it's crazy the amount of open problems in the privacy part of the, the, the deal that, uh, we don't even know how to deal with at this time. And they're all there. I mean, looking at Lindon for any security professional is, it'll definitely teach you something.
16:32Matt Coles And in fact, I mean, the, the confidentiality, integrity, and availability are prerequisites for privacy. Right? Confidentiality is obviously critically important, but availability and integrity do actually matter for many of these things. Interesting, interesting note, if anyone is paying attention, the, the, the first N in Linden and the R in Stride are mutually exclusive.
17:02Izar Tarandach When they both Mm-hmm. Mind blown. Okay.
17:10Matt Coles How do you have properties of repudiation and non-repudiation at the same time?
17:16Chris Romeo Hmm.
17:16Izar Tarandach No, wait.
17:17Matt Coles You have threats of— they're not mutually exclusive.
17:19Chris Romeo You certainly do because the threat on the privacy side is that there are some actions that they're, they're saying privacy advocates are saying there's some actions that you don't want to have recorded.
17:31Matt Coles That's right.
17:32Chris Romeo Because primarily because, you know, I mean, think of an example of a government tracking some type of citizen activity or something, or, or, or forcing a tech company to open their database to do business inside that, that country. Unfortunately, that happens a lot more than we'll ever know in the public world.
17:50Matt Coles Well, I think non-reputation is that you, you can, you can do something without having, uh, um, Attribution. I, I, I tend to look at non-reputation after having learned a little bit about it, uh, have from talking to Cam, but also Carnegie Mellon has a nice course on privacy engineering, uh, that I had an opportunity to take. And non-reputation I think is more along the lines of attribution about tying statements, you know, say political statements to an individual. You want to, you want to have plausible deniability to some extent.
18:26Izar Tarandach Yeah.
18:26Matt Coles So the way that I see non-repudiation is the way that I see it is I didn't do it and you can't prove that it was me.
18:35Chris Romeo Right.
18:35Matt Coles That's repudiation.
18:36Izar Tarandach And it was broken before I got there.
18:37Matt Coles Right.
18:37Chris Romeo That's repudiation. Not non-repudiation.
18:40Matt Coles Well, if you have non-repudiation, if you have a threat of non-repudiation, that means you can't, you can't do that. Yeah. Right. Whereas security, you want the threat is repudiation. And you need non-repudiation of the property.
18:54Chris Romeo This is why we've had all this tension between security and privacy all these years. Matt has pointed out the biggest challenge that we've had for the last two decades. I get it now. It's also, it's like, it's like I just, I took the red pill and I like, all of a sudden the Matrix is like, I'm seeing it being put together.
19:13Izar Tarandach That's a conference presentation right there. Yeah. And I would definitely come and watch it.
19:18Chris Romeo Wow. Bring popcorn.
19:19Izar Tarandach Better than that, just, uh, what does falsifiable mean? But that's another, for another time.
19:25Chris Romeo All right. The next one on this list is, I'm going to say one of the strange ones.
19:31Izar Tarandach Nope.
19:32Chris Romeo Yeah, it doesn't. So Matt, you spent a lot more time in the CVSS world and I can think.
19:38Izar Tarandach And thank you for that.
19:39Chris Romeo Yes.
19:39Matt Coles Thank you for that. CVSS is literally a severity scoring system. That is its job. It scores vulnerabilities. It does not score threats. You can use it for threats, but it is not a threat modeling approach. Certainly not a threat modeling approach. It is a tool that you may use when you find vulnerabilities in threat modeling, or if you want to simulate a vulnerability or weakness that you find in threat modeling, you may use it, but it itself is not a, not an approach to threat modeling.
20:12Izar Tarandach What happened to its lesser-known cousin, CWSS?
20:16Matt Coles It still exists, just not well funded or, or managed at the moment. Um, but it's, it does exist. CWSS, Covered Weakness Scoring System, is available for scoring the severity of weaknesses. It takes a more risk-based approach. Uh, which, so if we, if we think about pasta, our pasta discussion just recently, a little while ago, CWSS. Brings in some, some risk stuff in terms of business value and, and, uh, and likelihood of exploitation and other things that CVSS doesn't even attempt.
20:50Izar Tarandach I had so much hope for CWSS.
20:52Matt Coles So, and, and its, and its, uh, and its companion CWRAF, which is Common Risk Assessment Framework.
20:58Chris Romeo So, both of those exist, but— In looking at this though, the article doesn't give any rationalization for why the threat modeling methodology, like, they, like the, the author does with many of the other ones. And so I think we can just kind of cross this off and go, yeah, nice try.
21:14Matt Coles That's sort of, yeah. The final statement here is the CVSS method is often used in combination with other threat modeling methodologies, implying it is a threat modeling methodology. Not, it should say the CVSS method is used with threat modeling methodologies.
21:28Chris Romeo Yeah. To enhance them.
21:30Izar Tarandach Yeah. It came in as noise, probably.
21:32Chris Romeo Like, yeah. Let's look at the next one, attack trees. This is the one that's fascinating to me. I'm just, I'm still trying to grasp the concept and I'm still trying to figure out how to be a good attack tree person.
21:46Matt Coles Uh, it's cause and effect. What do you need to do as an attacker to get to a goal?
21:52Izar Tarandach I see it as the application of the Socratic method to application security.
21:58Chris Romeo Or the Five Whys.
22:00Izar Tarandach You can, you keep asking, what if this, what if this?
22:05Chris Romeo So it's the 5 whats, not the 5 whys, the 5 whats. Yeah.
22:11Matt Coles Or hows. The hows actually might be better.
22:13Izar Tarandach Hows, hows, hows is probably better.
22:15Chris Romeo Yeah.
22:15Matt Coles It's probably a, it's a path. It basically describes, it allows you to describe the chain of events, the tree of events that needs to occur to get from an entry point. To a goal, right? That's what an attack tree ultimately is. We talk about kill chains all the time, right? So if you understand the concept of kill chains, then you've got an attack tree. And as you, and this is the, this is the notion of if you identify a set of threats and then you sort of chain them together, can you get, you know, is there a path to get to that ultimate threat of data exfil? Or destruction of data or non-repudiation of a statement from a political enemy, right? I mean, these are the, what the attacker wants to do and how they get there.
23:04Izar Tarandach But that's the thing. You begin the process of building the attack tree from the end, from the objective.
23:10Matt Coles Right.
23:11Izar Tarandach So to me, it's always like that image of a smoky room, poorly bathed people, Going, dude, dude, what, what, what if giraffes all of a sudden stampede into the server room? Hey, put that code into the tree. Put it on the tree, man. Put it on the tree.
23:31Chris Romeo I'm not a fan of attack trees, I'm taking away here. No, no, no, no, no.
23:34Izar Tarandach I love them. I love them. Don't misunderstand me.
23:37Matt Coles I love them. They, they are more complex to build, right? They go hand in hand with, with, um, with DFDs or sequence diagrams, right? They're sort of another diagram that, that you should have in your tool belt when you're doing threat modeling.
23:49Chris Romeo Don't you need more knowledge though, just in looking at this? More—
23:52Izar Tarandach I think so.
23:52Chris Romeo More knowledge in your own head to create one.
23:56Matt Coles Yeah.
23:57Izar Tarandach Okay. So I think that being able to build a complete attack tree, and by complete, I mean as close as possible to complete, because it probably would be very hard to do a complete one. I think that you need knowledge of threat elicitation. That's probably much closer to a pen tester than to a blue team person, I think. So, for example, Jonathan, Jonathan has done, Jonathan Marcel has done a lot of work with attack trees, right? And I think that we had that discussion before that having an attack tree ready and handy when you're doing threat elicitation with whatever methodology of threat modeling you're doing. It's a hugely useful thing because you can put in front of a developer and say, when they say, oh, but who would do this? Or no, that's never going to happen. You can actually put it in front of them and, and, and show, no, this is the chain of events for something actually happening. My problem is that sometimes the, the, okay, the root, we know that the root of the tree is where you want to get is the X field is the. Uh, uh, RCE, but sometimes the, the, the, the leaves, they, they go so completely not out of proportion, but they get so specialized, right? It's like if you found this thing on a Thursday afternoon and you're turning to the east and it's Windows XP 2.0, whatever, then you get that thing started. And I can never like find where to stop the leaves. Of the attack tree.
25:39Matt Coles Well, I guess, you know, one way to look at this is when you ask that question of what, what can go wrong? And then you go, well, how do you, how do you get there?
25:48Chris Romeo Right?
25:49Matt Coles So if you, if you ask a sub-question of what can go wrong and then ask the next, the sub-question of, okay, so how do we, how does somebody make that, that happen? Right? So if, let's give an example here. You have a web server and a database. Well, attacker, attacker wants to get the, the credentials out of the database.
26:09Chris Romeo Okay.
26:10Matt Coles What, so that's what, that's what can go wrong. They can steal the credentials. How are they going to steal the credentials? Well, they have to break the front end. Okay. How do they break the front end? And then you can work your, your chain, right?
26:20Izar Tarandach Yeah.
26:21Matt Coles And each of those are potentially individual vulnerabilities that are leveraged or weaknesses that are leveraged or threats that exist. If you look at the MITRE ATT&CK framework as an example, That's where you might enter a bunch, have a bunch of the boxes that you would put into an attack tree and sort of build this path of, of, of how you would attack something. By the way, that gives you step 6 for, for BASTA if you're going to do that.
26:45Izar Tarandach But anyway, dude, BASTA is cool though. So let's keep going.
26:53Matt Coles It's supplementary. It's supplementary to other approaches for threat modeling.
26:57Chris Romeo I have never seen Persona Non Grata.
27:02Izar Tarandach That was my nickname in high school. He was the person.
27:05Chris Romeo He's always wanted to go to the parties, but was Persona Non Grata. Coming this summer on the WB.
27:13Matt Coles So is this the precursor to Elevation of Privilege?
27:18Chris Romeo No, I think it's like just, I think it's, it's, I see it as something just based on this description where You're thinking about the social engineering side of the threat modeling process, being like thinking about how people could break your stuff. So it's probably not applicable, like, when we're threat modeling a feature, because we wouldn't want to think about every possible persona. But if we're threat modeling as a security team that's looking at our organization, I could see this being beneficial.
27:47Izar Tarandach To me, it's more like not think like an attacker, but think who your attacker may be. So if I want to make sure that I am creating reasonable security, I have to know who I'm defending against and the level of reasonability going to be different if it's a nation state or if it's somebody who just wants to get credentials to sell on the dark web or whatever. Yeah. Well, that's where persona non grata comes to me and Actually, for people who are not familiar and are not looking at the article right now, it's basically a definition of personas and they are non grata because those are the personas that you don't want running around your system, right? So you define who they are, what they know, what resources they have at their disposal, how much motivation do they have to do what they're set out to do. So again, I would put it in the same category as CVSS in the terms of this article, It's a support mechanism. It's not a threat modeling methodology.
28:53Chris Romeo Okay. So let's keep going.
28:56Matt Coles To go with that is security cards.
29:00Chris Romeo I know about security cards. I have a copy of the privacy ones, the Linden Go that Kim signed for me behind me on my shelf. Never actually played them though. Then now this is a special edition. I would never take them out of the Fox because they've been signed by Dr. Woods. Uh, we've got Elevation of Privilege, uh, from, from the security side. We've got Linden Go on the privacy side. We've got OWASP Cornucopia, which I've never looked closely at enough. What's the, what's the benefit of these card games?
29:32Matt Coles Gamification?
29:33Izar Tarandach Yeah. Yeah. It's, I think that they offer structure where there is none. Right? Because it's a place to begin thinking about things. And sometimes in a fun environment, who doesn't like competition? Aha, I like it more than you. And, uh, uh, it just gets, gets people involved in the process, right?
29:55Matt Coles Security cards are more, are definitely, I mean, just looking at, if you read the descriptions, I've never, I haven't seen them myself, the security cards, but they're very adversaries focused. And of course, we, as we know from the Frontline Manifesto, that we should, um, be wary of focusing entirely on adversaries.
30:14Chris Romeo Yep.
30:15Matt Coles And so as a tool, it may be good for brainstorming.
30:20Izar Tarandach And what Matt just said goes for persona non grata as well.
30:24Chris Romeo Yeah. Yeah. I think it, I think the benefit of the card games is it gives you a fun way to connect with a team. That's new to threat modeling versus coming in and saying, everyone sit in your seats and prepare to be lectured to for 47 minutes on threat modeling and how awesome it is. Instead of that, it's, you're getting people engaged in the process. Yes, it may not be perfect from the categories and things, but you're, they're playing a game and we know as human beings, games are fun. That's, it, it's just fun to do that with a, another group of people. And in this case, you're playing a game and you're experiencing knowledge transfer. You're learning about threat modeling. So I think that's the real key. Even if the information is not perfect, I'm okay with that.
31:07Izar Tarandach And I'm going to go one step further and say that if you have, if you note down the results of this game of cards and write it between a system description and some sort of ranking of what you found, you got a threat model just as good as anything else that you would do with another methodology. It's just that depends on where your team is in the journey, right? Yeah.
31:36Matt Coles All right. So then, now we go to the next one.
31:38Chris Romeo I don't know. Let's see. We're 1, 2, 3, 4, 5, 6.
31:42Matt Coles Hybrid Threat Modeling Methodology is basically—
31:43Chris Romeo Yeah, we're, we're close enough to the end. We can, we can crank through all of them.
31:46Matt Coles This one actually will be pretty quick. It's basically take Persona Non Grata and take the security cards, add in security requirements and a, and a, and a and a system and a, and a model and iterate.
32:01Chris Romeo Hmm.
32:02Izar Tarandach So basically what you're telling me is that this one should be called SALAD.
32:08Matt Coles Well, he does, he does introduce another method called SQUARE, which is around Security Quality Requirements Engineering Method. Basically the thing that we do when we build systems and we look at functional and non-functional requirements and we define quality metrics and security requirements.
32:27Chris Romeo It feels so painful. All those words you're saying sound painful. Painful. We didn't hear—
32:33Izar Tarandach So the main, the main steps of the method are identify the system to be threat modeled. What are we building? Apply security cards on developer suggestions, remove unlikely persona non gratas, and summarize the results. Uh, what could go wrong? Use unconfirmed formal risk assessment method. What can we do about it? So yeah, it, it checks all the boxes.
32:55Chris Romeo Yeah, but it's, it's using a lot of words that feel like they're driven by government agencies and this just, once again, this just feels slow and yeah, waterfall-y to me when I hear these words used.
33:08Matt Coles The targeted characteristics of the method include no false positives, no overlooked threats. And consistent results.
33:16Izar Tarandach And so where's the fun of it?
33:17Chris Romeo Yeah, those are, those are not, okay, hold on. Stop the, stop the train, stop the bus, whatever we're riding in right now. That's not possible. Then if you're saying the targeted characteristics of the method, no false positives, no overlooked threats, and consistent results. So basically perfection. Like I would love that too. Like if I could just pay a small amount of money and get perfection in In threat modeling, I would never have another vulnerability, but this is, this is—
33:46Matt Coles Because you have infinite time, right? You have infinite time and free resources, right?
33:49Chris Romeo Yeah, with infinite time and infinite resources—
33:51Izar Tarandach Wait, why small amounts of money?
33:53Chris Romeo Will type out—
33:54Matt Coles Cost-effectiveness, the fourth one.
33:56Chris Romeo The entire works of Shakespeare on typewriters.
33:59Izar Tarandach But why small amounts of money, Chris? Consultants have to eat too.
34:04Chris Romeo Okay, I'll pay. I mean, if you could guarantee me no false positives and no overlooked threats, I think you could sell this thing for a billion dollars. To each Fortune 100 company and make $100 billion.
34:18Izar Tarandach Okay, next, quantitative threat modeling methods.
34:22Chris Romeo Quantitative.
34:23Izar Tarandach So this is ATT&CK trees plus strides plus CVSS.
34:27Chris Romeo Okay, this is just a made-up one. They had 11, they had 11, and they're like, it's gotta be 12. We said it was gonna be 12. All right, well, if we just throw a bunch of these things together, we can just call it a different thing.
34:40Matt Coles So the first step is to build component attack trees for the 5 threat categories of Stride. So basically it's, it's, it's a structured brainstorming approach. It's a, it's a structured brainstorming approach to Stride. How would you do spoof? How would, if spoofing is the goal, if spoofing is that, is that root of the, of the tree, how would you spoof? What would you spoof? How would you spoof it? Where, how would you go through your system and spoof it?
35:06Izar Tarandach Well, it's time for Spoof the Magic Dragon.
35:08Chris Romeo Yeah, let's just let that one go and just say, let's just, even though ChatGPT didn't exist, they had 11. Oh God. They had 11 and they put, they put it into a, they traveled in time, put it into ChatGPT and it popped out quantitative threat modeling will be the 12th method to make your list even. All right. Trike. Who would like to ride on a trike?
35:28Izar Tarandach So trike is interesting cuz I, I, the first time that I saw it was when, uh, we were doing some research for the, the Phish book. And I looked at the tool that they have in there and it was interesting. It just didn't work. So I went to see why it doesn't work because it's not supported anymore. Why is it not supported anymore? Because the people who created Trike, it's not writing about Trike anymore. So I don't know, every time that I see Trike, I'm like, wait, uh, yeah, yeah.
35:56Chris Romeo Develop the path. Look at the details that they're describing here. Each threat falls into one of two categories, elevations of privilege or denials of service.
36:08Izar Tarandach Yeah, it's, that's the thing that funneled Stripe down to its absolute minimum common denominators and created two categories out of things. And there was this tool, there was this spreadsheet, if I remember correctly, and you could get things working together. But at the end of the day, it was a lot of, uh, It was a lot of work. It felt what water, what Chris keeps saying. And, uh, it, uh, I didn't feel like the result was something that justified the, uh, the outcome.
36:42Matt Coles It also used attack trees. It uses the, it uses the threats. So the threats are put into categories, but each of those threats is the root of an attack tree. So again, you're, you're figuring out the path through the system of how you would exploit or cause a threat to occur. We do this today with Stride or the Four Question Framework. We just skip the middle steps in a structured way.
37:05Chris Romeo Yep. I'm not seeing a future where I'm converting to trike.
37:11Izar Tarandach Um, nope.
37:13Chris Romeo Nope.
37:14Matt Coles All right.
37:15Chris Romeo How about a quad?
37:16Matt Coles You go with a quad?
37:19Chris Romeo Riding quads. Exactly.
37:21Izar Tarandach Depends on the helmet.
37:23Chris Romeo Helmets. Who needs helmets?
37:26Izar Tarandach Um, this is threat modeling. Everybody needs help.
37:27Chris Romeo Everybody needs a helmet. Another t-shirt for you. Threat modeling, comma, everybody needs a helmet. Let's write that down. I know that's going on a shirt right there. TM, trademark, patent pending, all the other things I need to say. All right. So I didn't realize VAST was something that was created by another threat modeling company in the world. Visual, agile, and simple threat modeling.
37:55Matt Coles Yes, it's based on process flows.
37:59Izar Tarandach Uh, whatever those may be.
38:01Chris Romeo What is a process flow? Is that like a ladder diagram or something?
38:05Matt Coles It's, uh, well, I mean, so the more traditional process flows you might, might think of are Business Process Modeling Notation, BPMN, BPMN, where you basically describe call flows or how, how data, not necessarily how data flows, but how systems and how system components interact. That is different.
38:26Izar Tarandach Okay.
38:27Chris Romeo I just accidentally clicked on process flow modeling and this is what showed up. This is what showed up here. So this is, so I threat model this somehow.
38:43Matt Coles Yes.
38:43Chris Romeo Ah, I think the hydro-eat treater could, what if the diesel oil caught on fire?
38:51Izar Tarandach So you joke, you joke, but that's how Stuxnet happened.
38:54Chris Romeo Yeah.
38:54Izar Tarandach So it is from something like this.
38:58Matt Coles Again, this is, this is just a different type of, it's a different type of data flow, right? It's looking at how, how systems interact with each other. Call flow, call stacks are, are sort of an example, right? I have a function, that function calls out to the OS. Well, actually, it reads data from a file and then calls out to the OS with that data. I have a threat. I have a potential vulnerability because I have a weakness now of untrusted data going into a, into a call, right? But the call flow versus the data flow. They are different. Mm-hmm.
39:37Chris Romeo Okay. All right.
39:39Matt Coles There's value to, value to the process flow, whether or not the methodology is effective.
39:44Chris Romeo It feels more complex than it needs to be though. That's why, that's why I love data flow diagrams. I love simplistic data flow diagrams that only use process, data flow, data store, and external entities.
39:58Matt Coles Devil's advocate on this is there's structural information that, and contextual information you don't get in data flows. That we have to invent new things like who initiates a communication. So if data is flowing from point A to point B in one direction, that doesn't tell you anything about who actually initiated the communication. Process flow would.
40:20Izar Tarandach Yeah, but that's why we have annotated threat models as annotated, uh, data flow diagrams, as you pointed out to the Phish Book.
40:27Matt Coles Yes, that's right. Right. We had to invent something new in order to make it work. Because the process, a process, and by the way, sequence diagrams do the exact same thing.
40:36Chris Romeo Yeah. So those can be, those are, I think of those as ladder diagrams. Is that the same thing as sequence?
40:42Matt Coles What you're pretty close. Yeah. Okay.
40:46Chris Romeo All right. Brings us to our 12th and final threat modeling methodology, Octave. Octave is ancient, ancient, ancient tech.
40:57Izar Tarandach I, I, I heard of it. I have never seen it in the wild. Whoa.
41:01Chris Romeo It says created in 2003.
41:03Matt Coles 2003.
41:04Chris Romeo Yeah.
41:05Izar Tarandach Oh, that's why it's older than me.
41:10Matt Coles And, and it's, if you look at it, it, uh, it's, it's 2 step, it's 4 steps, 5 steps. 1, 2, 3, 3 phases, bunch of steps. And trying to get in here and to view this.
41:26Izar Tarandach Oh, it's asset-based.
41:27Matt Coles It has threat profiles, security requirements, risks, mitigation plans, technical vulnerabilities. You may look at, back at POSTA and OCTAVE, sort of precursor to, to that approach. I, I would, but I would suggest it's very similar, right? You're looking at assets, you're looking at requirements, you're looking at threats, you're looking at risk.
41:54Izar Tarandach But look, the, the, the article itself starts by saying that it's a risk-based strategic assessment and planning method. It points out that it's not per se a threat model thing.
42:07Chris Romeo Yep. But I can just write this one off because it's from 2003. I guess it's probably still used.
42:14Matt Coles It's probably still used in certain government areas.
42:20Chris Romeo It does not address technological risks. So it's really more of a risk management thing. That's what it says. I'm just reading from the article.
42:29Matt Coles It does not address— oh, it addresses organizational risks, not technological risks.
42:33Chris Romeo So it feels more like a risk management thing than a threat modeling thing.
42:43Matt Coles Well, let's see, in Fathering— Father Down—
42:46Izar Tarandach Father—
42:47Matt Coles Further down the paper, there's actually a handy table of what each of the methodologies helps with.
42:53Izar Tarandach Oh, look.
42:59Chris Romeo Yeah.
42:59Izar Tarandach Yeah. But it still goes with CVSS.
43:01Chris Romeo Directly contributes to risk management.
43:03Izar Tarandach So, yeah.
43:04Chris Romeo Which was my theory further up.
43:06Matt Coles CVSS is designed to be scalable.
43:09Chris Romeo Is anything from 2003 scalable?
43:14Izar Tarandach So all in all, I think that this article, which is like really, really quoted almost everywhere that threat modeling is brought up, I think that the article itself has a lot of value. It put together a lot of stuff that before nobody had bothered to.
43:30Chris Romeo Yeah.
43:30Izar Tarandach My problem is that people have been quoting this thing up and down the street for years now without giving it any critical thinking, without looking at the things that we just did. And just parroting it and going, oh, CBSS is a threat modeling, you know, and every time that I see that, I know two things. First, they're quoting from this article. Second, they have no clue what they're talking about. And if you look around, there is a lot of that stuff going on.
44:00Chris Romeo Yeah.
44:01Matt Coles Yeah.
44:01Chris Romeo I think that's a good place to, to wrap up this episode. Hopefully folks got some good information about some of these different things. Maybe you haven't heard of some of these, uh, different techniques before. Uh, we unpacked each one as we went through, gave some opinions, but we will be back next week with something else. I wish I knew what it was, but I'll know by next week.
44:25Izar Tarandach Well, you know where you're going, so we probably know where it's going.
44:28Chris Romeo I'll give you my thoughts on RSA for, for 2024, what I see. I'll give you some And then I'll try to make a case for why Matt and Izar need to go to RSA. That'll be my challenge. I'll bring you a bunch of t-shirts back. I'll see what cool t-shirts I can find, then I'll, I'll bring 'em back for you.
44:44Izar Tarandach So you won't find cooler t-shirts than what we have. Okay.
44:48Chris Romeo It's all the ones that you make that are the best.
44:50Izar Tarandach All right.
44:51Chris Romeo Thanks folks. Talk to you soon.
Transcript supplied by AssemblyAI. It may contain transcription errors.
Originally published as The Security Table. Part of the AI Security Table archive.
