50 minSeason 3, episode 10

Building the World's Largest Threat Model Library

With Chris Romeo, Izar Tarandach, and Matt Coles

Threat modeling & security architecture

Today we’re joined by Petra Vukmirovic. Petra, is the head of information security at Numan and co-leader of the Threat Model Library Project.

Episode title image: Building the World's Largest Threat Model Library

Listen to the conversation

About this episode

Today we’re joined by Petra Vukmirovic. Petra, is the head of information security at Numan and co-leader of the Threat Model Library Project. Petra shares her vision for creating a massive, structured dataset of crowdsourced threat models that could revolutionize how the cybersecurity community learns and shares threat modeling knowledge. We explore the complex challenges of convincing companies to share their threat models publicly, diving into concerns about legal liability, competitive advantage, and the fundamental tension between transparency and security risk. Listen along to learn more about this exciting project and its potential impact on the cybersecurity field.

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 · 50 min conversation

0:08Chris Romeo Hey folks, welcome to another episode of the Security Table. This is Chris Romeo. I'm joined by Isar and Matt, and we have a special guest with us today, Petra. And I would say we just jump right in and let Petra introduce herself just so we get some perspective on where you're coming from. So, Pesra, the floor is yours.

0:28Petra Vukmirovic Thank you, Chris. So, I'm, I'm Pesra, and I am, I've been in cybersecurity since about 2019. Right now, I'm in the role of head of information security at Numik. But because I'm also super passionate about threat modeling, I'm someone with a ton of ideas and not so much time. So I'm also doing a fractional head of product role in DevArmor, where we're creating threat modeling with AI. And I'm also the co-leader of an OS project called Threat Model Library that I can give you a little intro into.

1:15Chris Romeo Okay. I do have another question first. So when did you get into threat modeling? Like, what was the kind of, What was your threat modeling origin story?

1:27Petra Vukmirovic Love the, love the question. So I got into threat modeling as soon as I started working in cybersecurity as a cyber intern. So when I was working in Glasswall as a cyber intern, Denis Cruz, do you know him, was my CISO at the time, and he got me pretty much drawn in straight away. We started doing a bunch of threat modeling, um, and I loved it because in my mind, the way I see it, it's a brainstorming session. It's like we're discussing all things that go, go wrong. We're solving puzzles and it's something that kind of puts me right in. And I think, yeah, that since then I never looked back.

2:08Matt Coles Okay.

2:10Chris Romeo Very cool. Well, let's, uh, let's get into the project. Love to hear some about how the project works.

2:16Petra Vukmirovic So what I'm trying to do is create a massive dataset pool of threat models. And there's a lot of reasons why I wanna go for that. One is that, uh, when I try to teach threat modeling to anyone, most of the time I get asked with the question, okay, can you provide some examples, especially specific to certain industries? And this is where it starts to be difficult. There is not a lot of examples out there. And I always think that the community would benefit from having this. The other thing is, even if there was examples, they're not all structured in the same way. They don't contain the same, um, content. And I really was keen to create a dataset that is structured so that everyone can learn and benefit from it. In a certain way that they know, okay, what's the structure of, you know, the scope of the threat model, and then what are the security components of it, like the threats and the controls. So I always wanted that to be like a very structured dataset and to actually work to, you know, all the structured stuff as well. And then thirdly, uh, since I've, you know, been playing a lot with LLMs, I do think the community would benefit a lot from having to, having to be able to fine-tune LLMs. And to do that, we need high-quality threat modeling data. And that is peer-reviewed, that is crowdsourced, and that is creating a structured format. So these are kind of the incentives behind the OS project. And I've introduced it in Barcelona on the Global AppSec, and it's been met with quite a lot of engagement so far. So quite excited to see where it goes.

4:10Izar Tarandach I just want to take a second and point out that this was a new security table record. It only took somebody 3 minutes and 33 seconds to say the word Evelyn.

4:23Petra Vukmirovic Yeah. Yeah.

4:24Chris Romeo Proceed.

4:25Petra Vukmirovic I fell into the trap.

4:27Matt Coles Proceed, proceed. So we're holding our comments.

4:32Chris Romeo So I guess, how, how do you envision this thing being structured from? Because I think about when you think about an archive of all the world's, it's like the Library of Greece or something, or, you know, all of the world's documents and whatnot.

4:49Matt Coles Alexandria.

4:50Chris Romeo Alexandria. I don't know. Come on. It's early in the morning here. It's like 4 o'clock in the morning. It's 4 o'clock in the morning on the East Coast. Come on. So, but how do you see this being structured? Like, what do you, what do you, what are you, what's your vision for how you're gonna lay this out so that somebody can make it easy for people to find the right thing?

5:07Petra Vukmirovic Well, I think that's a great question, and I think I haven't answered that in full yet. The way I think we should do it is we should start, you know, just how to start, and each threat model will have a scope that is set up and it has, you know, all the details about it. So one way to structure it is to then kind of segregate into categories. So, you know, potentially give it tags and, you know, each threat model will belong to a certain category. And then based on that, they can be structured per industry, they can be structured per complexity because, you know, it's not the same thing if you got a threat model of one microservice or, you know, 10 microservices all at once, because it's a different zoom in, right? You're not going to be looking so deep into detail if it's 10 microservices, or, you know, you're going to be focusing on high-level stuff. So these are some ideas, but it's really at that point I haven't kind of put that into play. I'm waiting for a couple of threat models to start coming in, and then I kind of want to do it then.

6:19Chris Romeo So, is this the idea that this is an open contribution? Like, it's like an open source project or open source repository where we, we were trying to get anybody that's willing to submit something to put a threat model into the repository to share with others?

6:34Petra Vukmirovic Yes. So, highly encouraged contribution, basically crowdsourced. Myself and Julian, who's my co-leader, we're gonna be doing the peer reviews to make sure there's high quality. We still wanna find a way to benchmark it. I think that's also one of the challenges, but we do want to get the community to get involved and publish their threat models, basically. So we want to be able to, have a high variety of threat models that are made of existing systems. So for example, open source systems like Metabase or, you know, supported by Backstage, Netflix, Dispatcher. There's a lot of open source software out there that could be threat modeled. That way we are also elevating the security of these, uh, open source software and systems. And then also additionally, uh, people can publish their own threat models. Um, and that's like a whole new rabbit hole that we can open up. But, um, yeah, that's kind of how I see it.

7:36Izar Tarandach So that rabbit hole there, let's go straight in. So the open source, uh, uh, aspect of the thing is, is great. I mean, I think that ever since I saw the chat model of, uh, the K80s environment by Trail of Bits, uh, it, it, it always gave me the impression that that would be the next thing for people to contribute to, to open source, uh, projects. But now comes the question, for the volume of data that you want, probably open source is not going to be enough. And you may end up with like too many contributions of the same open source thing. And now you have to decide which one is alternative and which one is complete and all that good stuff. But another thing that has happened in the threat modeling arena, sorry, scene in the last few months is Adam's new initiative together with My goodness.

8:38Chris Romeo Lauren. Lauren.

8:39Izar Tarandach Lauren.

8:39Chris Romeo Von Felder.

8:40Izar Tarandach Yep. Of sharing threat models, right? And that's sort of like running parallel with your project in a ways. And then there's the third track, which is this effort to create threat modeling shareable formats coming from CycloneDX. And I think it's hard for me at least to look at this as a perfect storm of things, but all three point at the same thing. They run parallel at the same objective of making threat models more accessible to people. And as someone who spends a lot of his day doing risk assessments for vendors and stuff like that, At the end of the day, being able to turn to a vendor and say, show me your threat model, would be extremely helpful, or at least would save me a lot of time. Right? So how do you see from the answers that you are getting so far? People moving in the direction of sharing, people saying, it's a good idea, but can't. It's like, are you seeing people donating kidneys or hearts here? What's the—

9:57Petra Vukmirovic I think it's a great question. I mean, on my presentation in Barcelona, I immediately got, um, hit with skepticism. It's like, why do you think anyone in their right mind would share their threat model? And, um, I just wanna mention this is not a new idea, right? So Adam started it recently, but I know ever since I was even, you know, working in Glasgow, Denis was always talking about it. And we in security do keep talking about how we wanna end security via obscurity, right? We wanna be transparent about our security posture, and that way we can force everyone to be more secure. And I think the reason to publish those threat models is exactly that. We wanna build trust. Like, if I'm a company, And I actually claim to my clients, my customers, that I am being secure. Well, I should put something behind those words and actually publish that threat model and show that I'm not just, you know, talking the talk, that I'm not just compliant, you know, ISO 27001 compliant and don't have just the SOC 2, but I actually do care about my security and I do actually make sure that You know, I put all my business logic flaws, uh, into place before I put something in production. And there is also baby steps towards that perfect utopic world where everyone publishes threat models and everything is secure. And I think that's also something I wanna talk about in, uh, DDC if I get there. And it's, you know, you can start small, you can, you don't have to publish your whole threat model first, but you can publish just a smaller scope. And the way people will see it is like, you know, you are actually working towards something. You're publishing your success stories. Have you mitigated a threat? And it builds trust in the community. It builds trust between your customers, and it helps kind of establish a credibility. And that's kind of where I think, you know, vendors can benefit from it. Besides, if they do publish a threat model, they can get second opinions. They can get security advice for free. You know?

12:11Chris Romeo We don't have opinions. Security people don't have opinions.

12:15Izar Tarandach Who is ever going to look at something and say, uh, no?

12:19Matt Coles More importantly, a threat model, I mean, ideally, if we're doing it right, threat models shouldn't be, there shouldn't be opinions, right? Threat models are pretty exact. Once you describe a system and you have a set of threats and you have Rules for those threats, it shouldn't be opinionated, uh, unless you're using an LLM, because now you have hallucinations and other, other variations to worry about.

12:43Izar Tarandach Oh, no, no, no, no, no. I had hallucinations way before LLMs. Thank you.

12:47Chris Romeo Well, let's not talk about the '60s.

12:49Petra Vukmirovic Come on.

12:49Matt Coles So, so threat models that is our dos may have issues that require some opinionation. Uh, but Threat models are generally rule-based. Right? And so, hopefully, there may be difference of opinion in mitigation techniques, but not necessarily the threats that exist.

13:09Petra Vukmirovic Yeah. And that's kind of where I'm meant as well because you can mitigate a certain threat in so many different ways.

13:16Matt Coles Right. So, one of the, I think one of the biggest challenges you're gonna face, as you're talking, I had my, like, this is fun hat. This is a great idea. On my other hand, oh my God, this is horrible. Um, thank you.

13:29Izar Tarandach I thought that I was the only one.

13:30Matt Coles Well, so I, I'm, I've been in the habit of channeling my, my corporate attorneys recently. Um, this is a huge liability problem for corporations, right? Here, here's my threat model, what we did and didn't identify, what we did and did not fix or chose not to address. And that brings a whole host of problems, right? So, So there's a limiting factor I think you're going to run into unless there's a forcing function. I just did a quick check. SSDF as an example suggests threat modeling as an activity, but doesn't require it. You know, it's not something that you publish. The CRA, the, uh, EU Cyber Resilience Act, uh, does require manufacturers to pub— to create and document threat models, but not for public consumption. Right? So there's a, there's a, a medium there, uh, or, or a midway point of the threat model gets created, documented, provided perhaps to regulators and/or, um, assessors, but not necessarily required to be made public. Although I, I guess it'll be up to how the CRA is actually implemented as to whether that, that information gets publicized or not. Uh, so what's your, I guess what's your, I won't say pitch because it's, it's not really a pitch, I guess. What's your, what's your response to—

14:51Izar Tarandach Vision.

14:52Matt Coles Well, I guess how in this project, I mean, OWASP has a lot, has a number of corporate members and folks who are from corporations. So I can, I can kind of understand the pushback here that you were getting initially. What is your, how do you want to win over the the lawyers and the engineering managers of corporations, oh my God, this is horrible, versus the marketing folks who probably want to, you know, make this known, oh, security is a differentiator, and how do we make, how do we push out that message?

15:24Petra Vukmirovic I think there's a few things. No, I completely agree with you, Matt, first of all. And I do also get that little devil on my shoulder sometimes telling me like, whoa, are you talking about? Like, why would anyone do that? Um, you know, and I do think there, this cannot just be, oh, let's publish everything. We do have to have a certain caution and make sure we're also collaborating with our legal teams and, and, you know, ensuring that we're not actually creating more risk than solving it. The way I would approach this is, first of all, you, you don't have to publish the whole threat model. You start with a narrow scope. And, and you publish, you know, what you have as a success story and you publish that and then you move forward, like move onwards to the next scope. So starting small, you know, not going immediately towards a completely open threat model, but just starting small and leading by example. And the other one would be to also think about, okay, can we anonymize this instead? In some ways, right? Can we anonymize it but still not take away the actual meaning of it? Like, can we introduce some sort of obfuscation or anonymization that will allow us to still publish the gist of it? Like, not to lose on the threats, not to lose on the community learning, not to lose on, you know, the trust that we're building, but at the same time, in a way, have a caveat to say we are protecting self-confidence. Information here and, you know, make it also okay to do that. So I think there is like a middle ground between complete openness and transparency, which ultimately in the perfect world we will be in. But there is, like you said, some caveats to this that we have to consider. But I'm curious if you guys have any thoughts on this. Is like, is there anything else that we could do?

17:25Izar Tarandach Oh my God, she turned the table. That's it. That's it. We're gone.

17:28Chris Romeo Oh my God.

17:29Petra Vukmirovic Oh yeah.

17:29Chris Romeo Run away, run away, run away. So let me, let me go first. So until you mentioned the words anonymize and obfuscate, I was in full-on skeptic mode of this idea. Having just, I mean, having worked in corporate America and knowing, to Matt's point, the risk tolerance of the average corporate lawyer would be, I don't know, what's the number? Slightly above zero, but like many, many decimal points out into—

17:56Matt Coles It goes to 11.

17:57Chris Romeo Right. Yeah. 11. Yeah. So basically, yeah. Until you mentioned anonymization, being able to somehow take an artifact and remove the identifying information about it, I was completely gonna say this is never gonna work. But that's an interesting idea to give people. Now you'd have to find a way to anonymize in such a way that somebody, an attacker, and let's kind of boil this down to the core, right? What's the ultimate argument against sharing a threat model is, the best attackers in the world, primarily nation states, threat model before they begin their process of trying to break something. This is step number one. And if we're feeding them information, we're making their job even easier and they're already good at what they do. But we don't want to give them more data, more information to be successful faster in the things that they're doing.

18:51Matt Coles Especially since you mentioned training LLMs, right? So you're, You're going to weaponize this information right off the bat.

18:59Izar Tarandach Well, I think that as Adam puts well, they're doing just fine without it. So it's not like we are all of a sudden showing people how to build a bomb. Everybody knows how to.

19:12Chris Romeo Yeah, but they don't know what the— they don't have the blueprint for the bomb. True.

19:15Izar Tarandach So that on the point of anonymizing and obfuscating, which I don't quite like, but close that. The, the, the third leg of the triad— so we spoke about the, the, the sharing threat model thing, we spoke about the project, then we have the, the format, right, the, the, the SQL on DX effort. One of the, the, the important values to me that links to what we're talking about here is that me, for example, for my whole threat modeling practice, I've been pushing for people to always use one template for the output so that no matter who's reading the threat model, they know where the information is at all times. They don't have to go parsing things in their heads. They know where things are. So across an organization, use the same template. The beauty, I think, of having a, for the lack of a better label, BOM for threat models, which is what we try to build there, is that in this context here, it's going to be very easy for us to decide which fields need obfuscation, which ones need anonymization, and which ones need to be left out of what you're publishing altogether. So we would be able to have a format that people could feed into, for example, Petra's project, that might be partial in how much information it brings. It could be a complete format. If it's a complete format, then on the output of it, you would be able to sanitize it and say, okay, I'm going to go up to this level of detail in the way that I output my threat model, right? So by moving in the direction of having one specific canonical format that we are able to use to express a threat model, we start being able to automate, or at least to, hate the word, but it's a good one, censor. The output of that threat model in the way that it goes forward in other directions or downstream or gets utilized or things like that.

21:20Matt Coles I'd like to challenge you on something here. So, and I say you, all of you. So, it's one thing to anonymize or obfuscate information, you know, names and places and and, you know, methods and things like that. It's another though to anonymize the function that is actually being threat modeled. Right? So, if I, if we were to publish a threat model for, I don't know, PyTM as an example, right? It would be relatively easy, even if you anonymize all of the information in it, to either recognize that it's PyTM, or know it's something like PyTM, you know, a system that does something because the construction of it requires that structure. Uh, and so you can't really anonymize that away. Otherwise, and, uh, and on top of that, if, if you do anonymize that information away, do you lose the value? What's the value of publishing the threat model? Because now it's not that threat model. It's not, it's not a threat model that exists anymore. It's a generic or an idealized version of something that does something similar.

22:33Chris Romeo But if the scope, Matt, if the scope is small enough, you could anonymize it and not be able to then determine what the bigger picture is. Like, you could say it's an authentication system. If you could isolate that from the bigger picture, I wouldn't be able to use the bigger picture to figure out it's a ridesharing app or something. And that's fine.

22:55Matt Coles Yeah, and that's fine. But that defeats the other purpose of community sharing of threat models for systems. That people can use to build trust. If that, if you anonymize away that rideshare app, have you really shared the rideshare app?

23:11Petra Vukmirovic I completely see what you mean, Matt. And, and there's still no perfect answer to this, but I think these are baby steps towards it. I think, you know, like Chris said, if you make the scope smaller, you know, the structure of the TM bond, then Um, we've been talking about and proposing has this thing called scope and scope definition, and that's where you can put all the details of what's in scope and what's not. And, and even, you know, that, you know, say that some details have been removed. However, we still don't have an answer in full to this. And I think, I mean, the best answer would be everyone should just be secure, but obviously that's, uh, easier said than done. Uh, So, so I do think it's, but I think it's just baby steps, right? Like, let's dream big, but let's, you know, work towards that in a small iterative, progressive way.

24:07Matt Coles And don't get me wrong. I, I, like I said, I think being a project is valuable. I'm not arguing that. It's just some of the points here.

24:14Chris Romeo Yeah, no, and I'm the same. I'm in the same spot, right? Like, I, I think it's, I think the idea is valuable. I'm just, it's kind of going through my mind. How are corporations going to respond to this, right? And I guess when you think about this, like, what's in it for the corporation? They don't care about educating the community. Let's just be honest. I'm just going to call it like it is. They don't. That has nothing to do with their bottom line, educating the community. And so the only selling point for them would be, could I somehow get more people to buy my stuff because they trust it more? And so I think that's really the angle that you'd have to go at for, for, you know, profit money-making entities to contribute.

24:58Izar Tarandach So the, the, there's a thing. If you turn back the clock to the episodes that we had way back when we talked about SBOM, was it good for? We had the thing where we were saying, okay, so people are going to publish their SBOMs and, and they're going to see that you're using LibXML2. But nobody's going to stand looking at two SBOMs and say, oh, these guys use the same, but these guys don't use the PBIX. I'm going to go with these guys.

25:21Chris Romeo Right.

25:22Izar Tarandach So, and then in the end, what we see today is if we look at tools, they started using the SBOMs to enrich the way that they do, for example, vulnerability ranking. A project like this. With the richness of detail that it would bring together with a format that can be interpreted by a lot of different tools, in my opinion, would also bring something like that, right? You would, for example, be able to say, oh, in the threat model, that component has been marked as being more important or more central. And all of a sudden, that scanner there in the end of the pipeline is saying that there's a CVE for a library in that component, so perhaps I should rank that vulnerability higher, right? Now, is this an off-product of people publishing their threat models? Not really, but it is a step in the direction of them building their threat models in that format. So in the sense of baby steps, if we give them the capability of, again, censoring their threat models and publishing them in that format with less detail or with the amount of detail that they are comfortable with. By already having the threat models in the format because it's usable to them, we create a small trinkle of people who would possibly be willing to put in their trust centers. And for those who are not familiar, trust centers is a place in your website that usually you have to— sorry, people have to request access to where they're going to find your certifications for whatever they're worth, your sometimes design documents. And now they're going to perhaps find your threat model in there. So it's still gated. It's still something that the corporation is able to say who's going to take a look at with all the NDAs that are good or not, but that over time we could create a wave of people willing to publish those kinds of details.

27:32Chris Romeo It's like a pen test report. It's like a pen test report, right? People share those, but you have to go through a process to— they don't publish that on the public website, the pen test report.

27:41Izar Tarandach Exactly.

27:42Chris Romeo They publish it in a way that you have to go through some steps to prove you're a customer or you have a vetted interest in it.

27:48Matt Coles But I think the important part is that it's a regular format and it has other benefits. So a pen test report is an output. I think what is, or you're saying with the TM-BOM format, a corporation can take it, make it part of their CI/CD pipeline, use it in their vulnerability management program, publish it in the same way that you would publish a pen test report for your trusted customers. And that regularity is beneficial regardless of whether it gets published to the open source project of, uh, Datastore.

28:20Chris Romeo Petra, what are your reactions to this, this dialogue that we've been going through?

28:24Petra Vukmirovic I, I completely agree with it because I think it's, it's, it's, you know, all in the right direction, right? Like, we can't just think that from one day to another we're gonna open everything and it's, it's gonna be happy days. We need to create the environment that will allow that. And like Iser was saying, the thing that would allow that is if we have a structured format, we incentivize people to do threat modeling, they will already have those readily available and at least they can start sharing it with the customers without NDAs. And then once they have like this whole library of threat models, they can start thinking, okay, which is the success story I can actually publish online and I can start, you know, throwing it out. But, you know, if we create the environment when this is a thing and where this kind of like is an acceptable goal to go towards to, then we can pave the way. So I think it's great. Everything that leads towards that direction is fantastic, I think.

29:26Izar Tarandach Now, both Chris and, and Matt show the bump on the road from the side of the, the, the law people. I have a different one. What do we do about developers? Sure. I mean, no, don't. In the manifesto, we talked about analysis paralysis. We talked about the hero track model there. We talked about a bunch of things that come from the developer side. But now basically we are asking them to take all their dirty laundry. And put it in the sun. And I can only imagine, actually, now that I think of it, there is a good side, there is a bad side. Because people are going to be so paralyzed by the idea of publishing this stuff that I can't imagine so much discussion happening over each item of the threat model that finally we are going to see people addressing that stuff. But on the other hand, I can see a lot of people saying, okay, are we done? Did we do a good job? Is this good enough for publishing? Are these the right details? And that's a level of education that I think that goes even beyond the difficulties that we have had so far. How do you think that should be addressed?

30:48Matt Coles Oh, okay.

30:49Izar Tarandach All of you.

30:51Chris Romeo Let's hear Petra's take on it first.

30:52Matt Coles Yeah, go ahead, Petra.

30:54Petra Vukmirovic So I think, I mean, you're right. Like just thinking about, is this publishable or not? It will raise a bunch of discussions. And I do think developers are the people that usually, as like statistically, just based on my experience though, are the ones who tend to actually wanna do the right thing, but they get pressurized from a product. They get pressurized from other teams to deliver fast. So, so they're unable to do it even though Most of the time, they're the ones who actually want to do the right thing and actually find it frustrating when they don't have the time to implement the right methods. So perhaps in one way, this will, you know, be kind of like a bit an obstacle. And on the other hand, it could also be a neighbor because maybe some of the, like a cohort of senior, I'm seeing this as senior and staff developers who like been around long enough to see breaches happen because of lack of threat modeling. I see them actually welcoming this idea and saying, yeah, like we really need to get our ducks in order and make sure that we create something that's worthy of publishing. So I think there's gonna be two cohorts here, and I think probably a way to manage this, um, you know, I think we, we are gonna have to cross that bridge once we get to it. Because I don't have all the answers right now, but that's kind of how I see it.

32:24Matt Coles I'd like to start small on this. I think, yeah, I think there's an interesting use case here. One of the challenges we, I think we face, especially in large organizations, is the exchange of information between development teams. And combine that with, well, we would have to combine this with good tooling options. 'Cause right now the tooling options are, they're okay. If you talk about like the drag and drop type tools or the threat modeling as code type tools, and now you have AI LLM generating random stuff. I could see something like this as a way of facilitating exchange between development teams. So if you have, especially if you have inner source components, being able to publish locally for other development groups, but also to exchange information or to, if you have a way of connecting TMBOMs together, you know, object A and object B have, or object A exists in two places, and now you can connect those, those threat models together. I think that could be interesting. And also if, especially if you're using, I'm sorry, especially if you have good tooling here, where the development teams are using code generators or using, I don't wanna say vibe coding systems, but I wanna say, I wanna say, oh, see, AI-assisted coding tools where as the AI is building the code on behalf of the developer, it could be updating a tmblr objects. I think that would be very powerful. I think, I think that'd be very powerful. That would also be a good selling factor for getting it into developers' hands. And if you can do that, then you can start to expand that scope, right? And maybe that's a way of cracking that nut.

34:28Izar Tarandach And actually, you raise a very good point that the copilot, to use a generic name, could be updating the TMBOM as it goes. But on the other hand, if you have the TMBOM when you start, it's one more guidance for the copilot to write the code.

34:45Matt Coles Yep.

34:46Izar Tarandach Right. So if you do things right and you do your threat modeling in the beginning when you're thinking about what could go wrong, the same way that you would give it to a developer and say, here be dragons. Now you can give it to your copilot.

35:01Matt Coles So Petra, here's an idea for your open source project. Build an MCP server.

35:09Izar Tarandach Wait, wait, wait, wait. 3503.

35:12Petra Vukmirovic Oops.

35:12Izar Tarandach Can somebody write it?

35:17Matt Coles Well, wait, I'll just pull up who's your favorite Vibe coder and we'll just code it right now.

35:22Chris Romeo It'll be done by lunchtime.

35:25Izar Tarandach Somebody call Claude. Somebody call Claude.

35:31Petra Vukmirovic I agree. We do need to make sure to kind of alleviate that obstacle that is our son to get our tools, uh, in the best shape possible. And this is where I'm also hoping that the project can help. Like, I really want the project to be something that any tool can use to update themselves. So completely all for that.

35:53Izar Tarandach Now let's talk a bit about that magic word that happens so much when we talk about projects. Let's talk about marketing. How are you planning to make this thing like public and push it forward? And now that you have at least one people in Australia, everyone in a gym somewhere listening to the project, how are you going to We have, we have, we have ones or tens right now.

36:20Chris Romeo Hey, don't get too cocky here. Come on. We, we, we are known all over the world of this block here.

36:28Izar Tarandach So how are we going to, how, how are you going to get people behind you?

36:34Petra Vukmirovic So, um, well, first of all, I want to make sure that I publicize about the purpose of this. Right. I want to make sure that I have the community behind the reason why this is good. That's like, for me, the number one reason I'm trying to promote it as much as I can on conferences. Like, you know, I've been to Barcelona. I'm hoping to go to DC. And the other thing is I, I will try to, again, have more ideas than time, but I will try to write blogs about it and potentially, you know, on LinkedIn, publicize it. I'm hoping to engage other people's help in this, right? Like if, if everyone believes in the, in the course, then other people will mention it, talk about it. I really wanna stay engaged as much as I can in all the discussions that are happening on Slack, in the OWASP Slack. And so these are, you know, some of the ways that I'm trying to kind of push it out there. I think what I wanna do is I wanna talk to the co-leaders again of Threat Modeling Cookbook. I've spoken to Zoe before and, you know, I wanna collaborate with them to try and get some of the threat models already there inside our project. And the other thing is I'm trying to engage also with TMC to get some hackathons going. Yeah, there you go. So I'm trying to, you know, attack it from multiple fronts. As much as I can within the hours of the day that I've got. But I'm also hoping to kind of scale up my efforts by just making sure that everyone's on board with the reason why this could be good. And then help, you know, other people helping out with this marketing by talking about it, by recommending it.

38:27Matt Coles Have you thought about moving beyond OWASP? So OWASP is a great organization, but I think what you're looking for is you're looking for feeder projects. OpenSSL as an example, and the Linux Foundation has done a lot about securing open source code. And if you really want to build a share of content, partnering with OpenSSL might be a reasonable choice, right? Especially where they do have threat modeling is part of the process of, as maintainers of open source. Should be part of the process for projects. They do spend a lot of time on automation because obviously open source developers don't have large corporate infrastructure available. But anything that could be done cheaply, freely, or otherwise might be a good way to start bootstrapping, right? There's hundreds of thousands of open source projects that could, you know, start small and feed So something, I don't know if that's something you were considering already, but.

39:35Izar Tarandach If only we knew somebody at SafeCode who could start taking this to corporations, but yeah. I wonder. I wonder, I wonder. But one thing that I wanted to ask you, actually, on the side of this whole effort, would be to, again, think about those poor developers and do something practical in terms of teaching them how to get the track models that they may or may not have already and translate to this effort and talk about the format and explain the rationale behind the format and all that. Because at some point, I think that you are going to hit the wall of a lot of people saying, that makes perfect sense. How do I do it?

40:19Matt Coles When is PyTM adding this feature?

40:22Izar Tarandach So Threadragon is working on it and working very, very strong on it. Amazing team again. And I believe that they are very, very close to having something that's useful. And PyTM has a bunch of issues that need to be closed. By the way, Matt, I'm going to need a review by the end of the week.

40:45Chris Romeo And— Don't promise, folks. This is actually a podcast. This is a podcast. This is not just— Matt and these are catching up.

40:53Izar Tarandach And after that, we are going to go strong into it. But again, it needs to be said that Petra is doing a great job with the format that works for the project. And that's pushing forward the work that we had been doing on the TMBall, which was a bit stagnated. So now we have something practical that's pushing that forward. And it's going to be a bit of a back and forth until we find the right place to park, right? But yeah, we hope to see more tools adopting the format real soon now.

41:27Petra Vukmirovic Absolutely. Like, you know, internally, um, the debugger is going to be outputting that format. I'm really hoping all tools can adapt it so that it doesn't matter which tool you use, but that there is tools available, uh, you know, open source, not open source. And what I'm hoping is to, you know, I'm super happy that ThreatDragon got involved so quickly and so soon. So I'm like, that's super exciting. And I can't wait for that because I'm going to be testing it myself. And yeah, absolutely right. We just need to make it as easy as possible for anyone to do a threat model in that format and push it out there.

42:11Matt Coles So just, uh, for those who are not familiar with the format, uh, is it, is it methodology agnostic? In other words, STRIDE versus something else versus 4 questions versus whatever, or because that would be another limiting factor, right? We have lots of ways of doing threat modeling and, and you mentioned security, right? We should mention, of course, or keep in mind, of course, the privacy threat modeling is also a thing. Um, and there's other sort of other types of threat modeling that can be done. So, um, presumably the intent of both TMBOM, for those who aren't familiar, and the project that you're leading, Petra, are tool and methodology agnostic and not just security focused.

43:02Petra Vukmirovic I, I totally agree. Um, So I have tried to make it methodology agnostic, but I have encountered, um, people saying that it's not entirely methodology agnostic. So I guess the, as he is scratching his neck, um, so the tell, he has it. I think, you know, my purest intention is to make it agnostic to like, you know, specifically Stride or specifically FASTA, but I am taking pages from those books, right? So, um, it's, it's very difficult to say, yes, it's completely methodology agnostic, but that was the intention. And, and it's not like a, you know, Stride-based, uh, or FASTA. There's no elements to, to kind of directly say that, but there is again pages from those books that show that there is some of it inside it. And absolutely privacy, yes, 100%. The schema is tailored to account for number of records that have, that are in a certain database or something. There is, you know, also risk assessment that happens afterwards. And I published a specification that kind of helps people understand the impact. And one of the things is regulatory impact, which is the privacy impact. So I have tried to push in some of the privacy aspects as well. There's also data categorization, like, you know, is this PII or is it, you know, uh, sensitive data in other ways, like, uh, you know, medical records or is it financial data? So there is some data categorization there as well. So yes, there is an attempt to make it beyond security and methodology agnostic. How successful I am in that, I'll let you guys be the judge of it.

45:04Chris Romeo Oh, we will.

45:04Izar Tarandach Don't worry.

45:05Chris Romeo We will.

45:05Izar Tarandach Yeah. Yeah.

45:06Matt Coles The danger, the danger is inventing your own, right? Not, it's not a danger, but forcing a particular approach as opposed to, you know, allowing for, and I guess The TM-BOM has to support this agnosticism.

45:21Izar Tarandach No, um, what, what important thing needs to be said that, that being agnostic doesn't mean rejecting stuff. So having a JSON structure that as one of the fields has stride category shouldn't bother anybody, right? Because you decide if you're going to use it or not. So one of the first formats that we, we discussed back in the day and that I shared with Petra was the JSON representation of PyTM, which is basically just getting everything and dumping into one big JSON structure, right? The format that's in use right now does have a lot of options, does have a lot of attributes that you can decide to have or not. So if you prefer one methodology over another, it's not saying don't use it, it's just saying, sure, why not, but not everybody's going to use The same fields, the, the, sorry, not the same fields the same way. Everybody's going to use the same field the same way, but not everybody's going to use all the fields.

46:18Chris Romeo All the fields. Yeah.

46:19Matt Coles Right.

46:19Izar Tarandach And another aspect of that is that if you are working with one of the open source tools and use that format, and eventually you grow up to something like the Vici or some other good tool, you can just use that format to pull everything in and not lose the, the work that you did in the, the small ones.

46:37Matt Coles 100%.

46:38Petra Vukmirovic Exactly. And, and that is so cool about it because if you've done already so much work on your threat models and then you wire a new tool and you're actually, you know, progressing because maybe you're buying a Bento tool, something that's going to help you streamline it, it's great if you could just import all your threat models, right? Like you don't have to start from scratch because you've done so much work and effort and you probably maybe already have some metrics based on it. You're covering, you know, your coverage of, you know, have you threat modeled your critical applications? And if you have a way to just transfer it from one tool or another, or if you don't like the tool that you acquired after a year, you can choose another one, and you can just easily transfer from one to another. You're not limited by any means.

47:22Chris Romeo Yeah, definitely. That's good stuff. So Petra, I think a great way to conclude our conversation would be, let's come all the way back around to the beginning and, just give us a, give us like the 30-second elevator pitch on why people should get involved with this project.

47:38Petra Vukmirovic No pressure. Okay.

47:40Matt Coles Uh, there's no pressure and none so far, right?

47:44Petra Vukmirovic So I think we need to work towards transparency in full and ending security by obscurity, but we're not gonna get there in one day, we need to work towards that. And one of the baby steps towards that would be, let's create as much threat models as we can. Let's create all the context that needs to be there to achieve that one day. And I think one of the ways is to, you know, publish whatever you wanna publish from your threat models, or just publish an imaginary application, or, you know, just exercise threat modeling. Sharpen your skills, all of that you can do and push to the threat model library, and then everyone else will benefit. You'll see how other people do it. You will learn from them. We will all learn from each other, and we can also sharpen our tools and make them better, and that way make all the postures of our companies more secure.

48:43Chris Romeo Excellent. Well, Petra, thank you for, for telling it, you know, explaining the project to us and, and going through these discussions. We're certainly all going to be rooting for the project and helping in any way that we can. And we'll encourage our, our listeners out there to check out the project and see how they can get involved as well. So all four of them, yes, all four of the listeners, everybody's got to show up, okay, if we're going to make a movement here. All right, well, Petra, thank you for joining the Security Table, and we, uh, we wish you, uh, great luck in this project going forward.

49:13Petra Vukmirovic And thanks so much for having me. This was great fun. I really appreciate I appreciate it, and I appreciate your support to the project, and I love all the debates and discussions and the skepticism because, you know, just talking about these things is progression.

49:27Matt Coles There'll be more conversation when you're done, when you're gone.

Transcript supplied by AssemblyAI. It may contain transcription errors.

Originally published as The Security Table. Part of the AI Security Table archive.