Allie Howe: Hey everyone, welcome to another episode of the Insecure Agents Podcast here at Blackhat. Today we have Mackenzie Jackson here with us from Aikido Security. He is a developer, advocate and security researcher there. Mackenzie, would you like to introduce yourself?

Mackenzie Jackson: Yeah, well, you did such a great job. I don't have much more to say, but yeah, I'm Mackenzie, security researcher, advocate, as you said, yeah, working with Aikido. I'm passionate about code security and discovering vulnerabilities, finding malware, all kinds of fun stuff.

Allie Howe: Awesome. I have used Aikido myself for vulnerability management. I think it's a really like easy to use tool. It's easy to get spun up. And when I hear things about Aikido is the auto ignore feature.

Mackenzie Jackson: Yeah, it just, I think I did a video about it I posted on Twitter actually a while ago where I compared Dependabot's findings to Aikido's and Aikido was able to remove like half of the vulnerabilities that depended about found. So what that translates to is half the work for a security team to go in there and patch vulnerabilities for something like, you know, SOC 2 what they have to do. It's mandatory. For for sure. And I and I think it's kind of comes back to how Aikido was created and founded it. Like our CEO Willem, this is his fourth SaaS companies on three exits. He's always been the CTO, always been the technical founder and like, this is just what he's always had to deal with.

And so when he built a product for security, it was kind of like, I need to build something because the noise is just too crazy. And developers hate, hate, hate doing that. And I went through the the findings. I what Aikido's superpower. I always say what we're really passionate about is making sure that when we give someone an alert, we have as much as possible validated that it's a real alert. So that means like going to your dependencies conversation, it's like is OK, you're using a vulnerable dependency. Are you using it in production? And also are you using the function that makes that vulnerable going through those steps?

And then if not, then you it's not reachable, it's not a target. So, so move on. And I think that kind of shines through everything that we do. We're like very, very passionate about making sure we get as false positives down.

Allie Howe: Amazing. I am. I saw that the announcement from Claude Code and it was yesterday or last night something about they are going to offer vulnerability scanning as a part of Claude Code. Now there was some reaction I think from the community where a few people are like what does this mean for vulnerability scanners? Are they just kind of like dead in the water now that this is part of and neatly offered within like a coding agent? I don't think so. My take was no, just because like the moat has never been finding vulnerabilities, it's been how to prioritize them.

Mackenzie Jackson: That's this is. Yeah. What you say is is so true. And I say this a lot too, is yeah. Finding vulnerabilities isn't the challenge anymore. It used to be. It used to be. But now we're, we, we pretty much know about all the vulnerabilities, the steps after that is them saying, OK, is, is this actually vulnerable and, and reachable? And can you target? And then it's how do I prioritize all of this? So what is the most important to do? Because I don't know if you've looked at many security dashboards, but very few of them have no issues on them, right? So we're, we're, we've got a lot to kind of get through and areas like that.

So I don't think that that's, that is going to be be the issue. And I we need to get to a point where these AI tools are, are doing this, that are scanning for vulnerabilities after they produce code, because at the moment it's kind of crazy. It's the Wild West out there. The other thing that I would say though, is when it comes to these AI tools generating code, they're often they're, they're, they're not generating the entire application, right? There's tools that claim that they can do that, but we were talking about Claude Code and stuff. You're, you're, you're getting it to perform a task.

You're getting its agent to perform a task, write some code in specific areas for you. When it comes to vulnerabilities, it's actually the greater context that you need to look at because you could have an injection vulnerability where you're taking untrusted data and then you're passing it through all of these different functions before finally using it in a query or something. And we saw like this is the SQL injection from MOVEit, which is a file transfer tool. And they had a big SQL injection vulnerability. And when you actually traced it back, it's because this untrusted data went through so many different endpoints and systems and functions before it was finally used that without that context, it's very difficult to just kind of find and discover those.

So it's I, I want to be clear, I feel like that's a great step for Claude Code and a very necessary 1, and I hope everything else follows, but this isn't going to solve security 100.

Allie Howe: Percent, yeah. It's just another like tool to use for sure. And and we need more of them and like bring it on that ABS absolutely.

Mackenzie Jackson: And I do believe we'll get to a point where AI generated code is kind of mostly secure like. But the the challenge is, and when you look at the research and academic papers on this topic, they're actually, if you could take benchmarks from GPT 3. 5 to GPT 4, for example, it's getting more insecure in the code that it generates. And at first it doesn't make any sense. But the reason why it happens is because they're being trained on larger data sets, not curated data sets like we you can't, you know, find you. So you're getting more and more, more and more code bases out there.

Take a dot, throw it out a random repository and tell me if the code inside is good or not. So we're not, we're, we're not going to solve this by making the training of the models better, at least not yet. So we can solve this by layering tools on there. And yeah, props to Claude for for for taking initiative and doing that's needed.

Allie Howe: For sure. And also how your code is actually deployed and where it's deployed counts too. So it's like, yes, maybe the code that Claude Code generates might be like more secure now that we have this. And maybe it gets to the point where it's mostly secure like you were saying. But how your applications actually sitting in a cloud and their architecture that influences the the different vulnerabilities that are present in your application A.

Mackenzie Jackson: Hundred 100% and I mean looking at creating perfect code as a one out of 20 things that you need to do to be secure. Making sure your endpoints don't have vulnerabilities that they're making sure you're a cloud infrastructure doesn't have excessive permissions and areas like that. Making sure that your YAML files when you spin up your infrastructure isn't creating vulnerabilities that way. Making sure your dependencies are secure. All of this kind of comes into it and this needs to be holistic is one of the things that I love about Aikido is I've previously worked for point solutions, meaning they are definitely the best at what they do, but they only do one thing and there's only so much you can do with limited context.

And when you take something like when you're looking at your your code and the cloud and the dependencies, you can actually bring all of this data together to create. That's how we know if a dependency is reachable or not because we have contexts that we can look at and we can do it. So when we had this holistic view of not only looking at your code and your dependencies, but we're looking inside your repositories for things like secrets, we're looking at your cloud infrastructure, we're looking at your YAML files. You know, we have a firewall in there just for, to, to kind of at the last moment, we have things like hardened images and all this.

And it just, it not only it's like nice to have as gets me to say lots of acronyms and it's fun, but it what it really does is give us more context to improve each of those individual points.

Allie Howe: Amazing. It's like it's really cool that Aikido not only helps with SAST vulnerabilities and SCA, but there's also like 8 other scanners right that you offer.

Mackenzie Jackson: Yeah, exactly. It's kind of we, we, we call ourselves like everything you need from code to cloud, right. So from the first line of code is like an IDE plugin. We're looking at your, we're, we're looking to make sure your code you're writing in real time is secure, but then all the way up into your, to your runtime environments, your cloud accounts, the firewall, the actual application. So like the, the whole, the whole journey along that everything that you need to be to be secure is there. And that just gives you so much better visibility. And and as I said, just more more data points into actually understanding how all of this complex puzzle fits together and and is it secure?

Allie Howe: Absolutely. Are you using AI at all today to help triage and prioritize one of those?

Mackenzie Jackson: Of course you can't, you can't not talk about AI. We have some amazing AI functions and AI teams that that have joined Aikido. It's really, it's really exciting. What's what's going on in there. Some things I'm not allowed to talk about just yet, but we have a lot. We have a lot that we're doing. So one of the big parts is, is triaging it. We've already touched on the fact that finding a vulnerability isn't, isn't a challenge anymore. It's what is my most critical vulnerabilities and how our triage works is it takes the vulnerabilities that are found and then it applies it in a greater context of your entire application across all of the spectrums.

So we're we're looking at areas. So for example, if you have a SAST vulnerability, but also a misconfigured YAML that's making your cloud, you know, you're just give me some vulnerabilities in your cloud, then we can accelerate this because it's, it's increased the risk of that being exploited in the wild. So areas like that where we can, because we look at the whole context, our auto triage really can be very powerful and making sure that not only are you dealing with what's critical, but you're actually dealing with what's critical and what's exploitable right now in this moment and dealing with that. And we also have a whole bunch of really cool features like Autofix, where if you actually find a SAST vulnerability, we can fix it for you.

And we do this in a like a very cool, cool way where we create sandbox from, from the code and then run it through multiple simulations and ensure that what we're suggesting is secure as well, you know, so there's lots of cool stuff happening. And yeah, we're also slowly moving into that. Well, not slowly, very quickly moving into code quality. So one of the things that I'm allowed to mention, I'm just not allowed to go into too much detail yet, but coming out soon, it's going to be some really cool, cool code quality features. So not only just making sure that you're secure, but making sure it's efficient, that you're following stick, good standards, good coding practices, that it's readable, all these things which are really important when everyone's using different AI models as well.

Allie Howe: Yes, 100%. Thanks for sharing that with us in our audience. We appreciate that. That's really exciting. And I think I noticed that too, when I use different coding tools and kind of between like my code, the models code and getting it to be seamless and readable and work together, there's different coding styles at play, yeah.

Mackenzie Jackson: Yeah, yeah, it's. Really gross we're getting people call like AI slop. Yeah, AI slop. Yeah, yeah, yeah. And the way AI codes too, is it's it's like layered, like you get to do something and then it doesn't work or throws an error. So you get to do much. And then like the actual code that you often end up with is gigantic because it's just kind of like keeps adding more and still it's like finally. And then what could have been so much more efficient is actually just just gigantic monolith of code that no one could ever like, comprehend or understand because there's just so much going on so.

Allie Howe: Much going on, it's like just going to blow out the context. Window it's just. Like not even I can understand it.

Mackenzie Jackson: Yeah, yeah. It was like when you first started coding, when I first started coding. This is going to be one of those like I'm so old moments, but you know, it was, it wasn't AI or Stack Overflow. And you'd just copy and paste from Stack Overflow and you'd like just copy this chunk of code that you had no idea what it did, but it worked. So you just kind of like put it in there. Not that I ever did that. Yeah, yeah, yeah. I'm not No, no, no. I won't be sure. I definitely wouldn't do that now. But you know, we can start. Yeah. You try to get something to work, you know, and that's the same with AI.

You just copy this great chunk and you put it in there and it's like, I don't know what it does, but it works, you know?

Allie Howe: Yeah, it's the same that we used to have a stack overflow all the time. You go stack overflow. I don't mean is the website still like alive, like, I don't know. I hope it's alive and well. I used it all the time. Yeah, yeah.

Mackenzie Jackson: It's still it's it's still there. They're they're going through some transitions. We'll have to see their their business model I don't think will survive the the revolution in the same way AI revolution. I think they're going to, but they've also been involved in helping train AI models in other areas. So there's definitely some pivoting going on in there and it will be interesting to see if it does still remain really relevant. I think it will have to go through a change. But yeah, it was such a lifesaver. I don't think I would have gone through uni without Stack Overflow.

Allie Howe: It was, it was like a lifesaver, but it was also character building. Like I remember you would like post something or someone post something and then like some, I don't know, really senior engineer or somebody that was totally sworn up and was valid and then saying whatever they said. But they would make you feel dumb for having asked the question. And like AI doesn't do that. Like like I feel like there's always junior developers out there that are missing out on that experience. Not that maybe they need to, but.

Mackenzie Jackson: It's way too positive most of the. Time. Yeah, yeah.

Allie Howe: This is a great idea for something else that. Plays you up and it's like, yeah, this is a good idea, but here's what I was thinking.

Mackenzie Jackson: Yeah, yeah, yeah. We're going to have a whole bunch of very, very propped up junior developers.

Allie Howe: Yeah, out of there. Yeah, Unless, I don't know, maybe you connected to like Grok or something and like unhinged mode or yeah. Yeah, OK. Yeah. I. Don't know. Yeah. Or you could get on like the GPT marketplace and things you can like buy or not just use different models like one of those specifically like so that GPT isn't so polite and can actually be, you know, somewhat honest.

Mackenzie Jackson: Oh, interesting. Yeah. Yeah.

Allie Howe: So I'm sure you see tons of different code samples from the developers that you help secure through Aikido, all your customers. What's something you've seen that surprised you lately?

Mackenzie Jackson: I've always been surprised at the amount of SQL injection and other type injection vulnerabilities that that we still see. And so we did some research and if we kind of take some related vulnerabilities like SQL injection, no SQL injection, command injection, path traversals, those types of things, which, which have been around since the Internet, are, you know, we still see that this makes up around about 15% of all the vulnerabilities that we're finding. And why I find that so shocking is like in particular with SQL injection, which makes up for around about 7 1/2% of all vulnerabilities that we, we're finding is that the solution to it is so well known.

It hasn't really changed, just kind of prepared statements, parameterizing your queries. And we're still struggling with even these, these, these basic things. And, and it's important to note too, that these AI models being trained on code that uses all that as well. So it keeps slipping in. So I'm always a little bit surprised at, at that. And I, and I think that it just kind of it, it, it just comes down to the, to the fact of, I think people misunderstanding the, the vulnerabilities. And, and when people think about SQL injection, I think they typically think that it's going to be something like really, really simple, like you have a login page that you can use.

But actually, as I touched on before, it's kind of when you go through all these motions and you lose track of where your untrusted data is and where it's being introduced and they kind of keeps popping up. So that's kind of something that, that that surprises me a little bit that it's, we're not kind of like seeing a vulnerability, figuring out how to solve it and then never seeing it again. We're just kind of like we're finding more stuff this this new categories of vulnerabilities, and we're still struggling so much with the old stuff.

Allie Howe: Yeah, I think that's sort of a general theme I've seen lately of like whatever problem you're struggling with, AI is just going to amplify that and make it worse.

Mackenzie Jackson: Yeah, definitely. And and like if nothing else like we, we, it's, it's going to increase, right. So the output is just like through the roof at the moment and what people are doing and the applications that are being spun up at, at record times. That also means that you're the amount of vulnerabilities in everything that you have is going to be spinning up in equal measures. So it's going to get harder and harder to kind of look at these dashboards and actually even figure out what to do. That's why remediation and prioritization and removal of false positives is so important now more than ever. Because Once Upon a time it was manageable to have a few false positives come through, right?

And you could tune it so that you know you have a manageable amount. Now it's kind of we really need everything to be pretty on point if we got any hope of actually dealing with what's important.

Allie Howe: Absolutely. And I work with a lot of small startups as part of the work that I do as a like BC. So for startups and these small teams just do not have the bandwidth to manage this giant backlog.

Mackenzie Jackson: And absolutely not, yeah. And anyone that's been in those small teams too, you'd like, you're just dealing with, with, with, with so much, so much stuff to kind of go through and not understanding it. And the, and the challenge with security too, is that even if you do everything perfect, you can still be vulnerable to different things like with your dependencies or of other areas. So, you know, it's a, it's a continuous struggle and it's not something that you can do once and go, oh, I checked my code. It's all good. Like, off we go. Happy days. Don't need to do this again.

Allie Howe: Yeah, I'm not. Going to check it. For a few minutes, yeah.

Mackenzie Jackson: You know, it's something that you constantly need to do, even if you haven't changed anything. Unfortunately, our new vulnerabilities could be introduced.

Allie Howe: Oh absolutely. Yeah, new vulnerability on all the time and dependencies and also LLMs changes. People interact with them, not with different code deploys. And like you're mentioning before with the maybe the sequel injections aren't kind of surprising, but we think with agentic systems now we're connecting our databases to all sorts of different MCP tools, agents. I mean, who is writing the queries and what's writing the queries just just become exponential. So like having eyes into that super important. So tell me a little bit more about the research you're doing at Aikido that's powering all of the features that you have.

Mackenzie Jackson: Yeah. We got a really cool research team too. A lot of our research is kind of it's, it's, it's Jewel in the sense that it's kind of uses scanners, LLMS, but also kind of very experienced security researchers as well. And we do a lot, we have our database called Intel, which is where we find vulnerabilities that have never been disclosed are. So they don't have CVE numbers to use the technical terms attached to them, but they're still have big vulnerabilities. And we've found thousands of projects and very big projects that are have a lot of dependencies that have security vulnerabilities that have been fixed or patched but never disclosed.

So people using lower versions don't know. But the big research that we do is a lot of malware discovery. So we actually scan every single package on npm and PyPI and some other places for for malware that's sitting in there with in June alone we found 4000 malicious package versions just just on npm, PyPI mostly. And yeah, we've made some really, really cool discoveries. There's a cryptocurrency called Ripple. It's one of the largest cryptocurrencies in the world. We discovered that that was actually had a back door in it and the official SDK. Someone had managed to compromise a developer's account, sneak in some malware that stole everyone's wallet keys, which could have been.

Really catastrophic, particularly because this SDK was being used by kind of a lot of the and all of the big exchanges. So we had a pretty catastrophic effect from from that and a lot of other kind of like malicious packages that we've that we've found the we have a fun story of when we when we kind of we're watching Lazarus, which is a North Korean hacking APT. They were kind of creating malware and putting it on npm and we were kind of spying on the mess. I doing it because they didn't realize that we were actually watching watching them upload these new versions of their malware in real time.

So it's been pretty fun of what we've been doing at the Aikido research team.

Allie Howe: Oh my gosh, yeah. That's amazing. Is there anything else that you're working on at Aikido that's exciting to you right now?

Mackenzie Jackson: There's so much that we're working on that I think is really exciting. I kind of briefly mentioned a little bit of the code quality stuff that I'm allowed to say, but not really allowed to say that's that's going to be really, really cool. And yeah, there's some other things that I can't mention that are going to be that are going to be fantastic. But yeah, we're really excited. And what's been great at Aikido is we've actually had our, some teams come join the company that have been doing really amazing research, really cool, groundbreaking things, particularly in the space of AI and security. And we've been, yeah, integrating with them and, and, and there's going to be some very exciting developments coming out in the future.

I just can't quote it.

Allie Howe: When are they going to be announced or how can we follow and say?

Mackenzie Jackson: Well, code quality is coming out soon, OK. So you'll be able to see that. And then then you're just gonna have to, you're gonna have to. I'm gonna be a good boy. I'm gonna, I'm gonna stick to my my lines. And Mackenzie, don't say anything. So you're just gonna have to hold, hold on. But that will come out and they'll be very exciting things coming out in the future.

Allie Howe: I bet that's amazing. I think of one of their differentiator for Aikido too that I've noticed is you meet developers where they are. Yeah, especially with like the cursor integration being able to just scan code, the time of vibe, so to say, or is what it's generated.

Mackenzie Jackson: It's, it's so important because we have a great, we have an amazing dashboard, looks amazing if you want to do it, but a developer doesn't really want to have to go to a dashboard to see security, right? They've got their own priorities. So it's really important that we, we meet, we meet them where they are. So that might be in your IDE, like if you're using cursor or if you're using VS Code or Jetbrains or something else. But it also might be in your GitHub actions workflow or you know, in your CICD pipeline. So we can actually find a vulnerability there and then give you a comment with code to fix it immediately exactly where you are.

So you don't actually need to leave your, your spaces that you're used to, that you that you work in. And then, you know, obviously then be able to give managers, team team managers, security folks a holistic view as well on what they own is very, very important, but just be able to meet them where they are and create it as frictionless as possible. I mean, you know, we've touched on the simplicity of Aikido is that a lot of these dashboards really used to look like fighter jet pilot cockpits where you just you open it up and they're just trying to do so many things. But actually what do you need?

You need to know what is the vulnerability? Is it real? How, how do I fix it right? You know, like so and so that's some of the, some of the really cool things with with Aikido.

Allie Howe: And that all sounds so simple, but it's really not. It's.

Mackenzie Jackson: Yeah, I think it's, I mean, it's a little harder to do that because you have to really like focus on and yeah, there's a like a very, we very really add a new like feature or a button or anything this but like to we take this to the extreme. One of my favorite features and Aikido is something called breaking changes. And basically, if you have a dependency, how you solve a vulnerable dependency is you upgrade to the latest version? Yes, easy. Unless of course, that breaks your application.

Allie Howe: Which is so I need dollars QA and stuff. Oh my gosh. And you offer that that's. Awesome.

Mackenzie Jackson: Oh yeah, we have a breaking changes analysis which looks and checks to see like. We'll upgrade into this next version. Break your application.

Allie Howe: Oh, that's amazing.

Mackenzie Jackson: Yeah. And the the the feature is like so hidden because the like this is such a fight is that we don't want to add a new button. Save. You so much time, but it is there and it's amazing and it's good, but that's just like a yeah, a little a little tidbit from from we really take it extreme. Just making sure that everything you need is there, but it's it's as simple as possible.

Allie Howe: Wow, as a developer myself, that sounds like a great experience. So that's awesome. I I think some like security professionals that maybe haven't served time as a developer or sat close to entering, they don't have quite the empathy of like what tools they're assigning their teams to use.

Mackenzie Jackson: I mean to everyone has different priorities too, and I can imagine that it's frustrating on on all sides where you have security people that don't think that this is being taken seriously enough. You have developers that kind of keep getting false positives or keep getting talked about and they're trying to meet deadlines too. So I think it's just important that everyone has a good experience in this and we need the teams to be able to work together. And if, and I think in the, in the revolution that we're going through at the moment with AI, it's our, it's really going to help and be able to build these products that that can actually start really meaningfully solving some of these problems between developers and security teams.

Because you know, with a tool in place that helps you fix something it it all becomes so much easier.

Allie Howe: Amazing. Very cool. Is there anything we did not touch on today that you'd like to?

Mackenzie Jackson: I mean, I could do a four hour episode in chat.

Allie Howe: Strap in.

Mackenzie Jackson: No, no, no, I think you know, but I just, yeah, check out the check out the Aikido and what I was doing. And I mean that will our that will kind of do a better job than me chatting about it for for a while.

Allie Howe: Awesome. Yeah, It's free to try, Yeah.

Mackenzie Jackson: Does it There was going back to our our CEO, they had a couple of requirements in the in the company and one was that you have to be able to try the product for free with like no credit cards and you have to be able to buy it without talking to a salesperson. So we have plenty of sales people so and, and I'll chat to you endlessly, I promise. Yeah. But you know there there's just kind of like you it has to be an option where you can just bypass that and just go. I just. I just want to start.

Allie Howe: 100% Well, amazing. Thank you so much for coming out today. Really appreciate that. It was Mackenzie Jackson, everybody from Aikido, security developer experience and security researcher there. That wraps up this episode of Insecure Agents Podcast. We'll see you all next time.

Mackenzie Jackson: Thanks so much for having me.