Zero Trust or No Trust: Why Enterprise Security Leaders Need to Rethink Mainframe Risk

The most persistent myth in mainframe security isn't that the platform is vulnerable. It's that it's secure by default, no additional work required.

Daniel Newman and Broadcom’s Greg Lotko bring Philip Young, Director of Mainframe Penetration Testing at NetSPI, onto The Main Scoop to break down what actually puts enterprise mainframe environments at risk as they connect into hybrid cloud, APIs, and AI-driven workflows.

Young draws on more than a decade of hands-on penetration testing, including time at Visa and Wells Fargo, to walk through how integration layers, not the mainframe core itself, tend to introduce the most exposure. He details a real incident where a web application scraping mainframe transactions exposed access far beyond its intended scope, and another where an internet-facing mainframe exposed backend systems through an internal network connection nobody realized was reachable. Young also addresses where AI fits into both defense and attack, including a live example of an AI chatbot being socially engineered into handing over account access, and closes with the one assumption he wants every enterprise leader to challenge: trust nothing outside your own known environment.

Key Takeaways:

🔹 Integration layers, not the mainframe core itself, tend to introduce the most risk. Young cites a case where a web application scraping transactions exposed far more data than intended because the middleware team wasn't aware of the underlying access controls.

🔹 “Defense in depth” means every layer only secures its own domain. Young argues each system component should enforce its own access rules independently, so a compromise in the cloud doesn't cascade into full mainframe exposure.

🔹 Social engineering remains the most effective attack vector regardless of technical sophistication. Young points to a case where an AI chatbot was manipulated into changing account credentials, evidence that social engineering now extends to AI systems themselves, not just people.

🔹 AI tools are augmenting penetration testing, not replacing the practitioners doing it. Young describes NetSPI's approach as "person in the loop," where AI accelerates output but still requires expert review before results can be trusted.

🔹 Flat networks and assumed trust remain the biggest structural weaknesses. Young warns that mainframes straddling corporate and commercial networks become the weakest link attackers will find, regardless of how well either side is individually secured.

Mainframe security keeps getting treated as a solved problem because the platform has a reputation for resilience. Young's message across this conversation is that resilience only holds up if organizations keep doing the work, not because the platform does it for them.

Watch the full video at sixfivemedia.com, and be sure to subscribe to our YouTube channel so you never miss an episode.

IMPORTANT: Note to Uploader: VIDEO EMBED – added by uploader.

Disclaimer: Six Five Media is for information and entertainment purposes only. Over the course of this webcast, we may talk about companies that are publicly traded, and we may even reference that fact and their equity share price, but please do not take anything that we say as a recommendation about what you should do with your investment dollars. We are not investment advisors, and we ask that you do not treat us as such.

Philip Young:
I'm glad the narrative has shifted from we're replacing the mainframe to we are including the mainframe now in these hybrid environments, right? Because before the idea was, oh, we're going to lift and shift, right? And then that never happened. So now we're modernizing our mainframe platforms to be more connected in a connected world.

Greg Lotko: 

Hey folks, welcome to the next episode of The Main Scoop. I'm delighted to be here with my co-host, Daniel Newman. Good seeing you, Daniel.

Daniel Newman: 

Greg, always good to be with you. Smiling big, ear to ear, because I know that every time we get together, we have fun, we talk tech. This business that we're in, it's, it's happening. It's been a great couple of years. I guess tech never wasn't happening, but boy, is it happening now?

Greg Lotko: 

Well, the reality is it's always changing, which makes it always fun, right? It's the variety of life. So our topic for today, I want to talk about mainframe security, and there's a lot of myths out there. There's the reality of where we are today, what a modern risk is to the platform, thinking about enterprises in general and what that attack surface looks like. But for me, the biggest myth about mainframe has always been that it's just secure on its own, that you don't have to do anything, that because it's mainframe, it's secure. And I tell people all the time, no, no, no, no, no, no. It is the most securable platform on the planet, but you actually have to do the work. You have to implement and use good hygiene and take advantage of the technologies. What do you think, Daniel? Is that what you hear while you're out there?

Daniel Newman: 

Well, I mean, it's not going to do the work by itself. Maybe it will soon, right? We all heard about the big Claude project mythos that went out there and found, you know, some incredible number of vulnerabilities across open source software, across the web, and it was able to call them out, identify them, fix them. Again, a lot of hype. My joke about Dario from Anthropic is he's moved more market cap to his private value by telling these stories of doom and fear. But one of the things that I did like about what he's doing is he's bringing attention back. And I do think at times security gets put on the back burner. People look at it like an insurance policy, like, yeah, we did as much as we had to do to mitigate the risk, but no more. And I think what AI is creating is a bigger attack surface. It is creating a set of tools that enable black hats and problem children and those that want to create issues, more toys to do their worst. And so I think this attention is really good. And whether it's the mainframe, whether it's data centers, whether it's your own PC or your own devices, I think security is going to be one of the biggest trend lines of the next 5 to 10 years. I mean, it should be forever. But this 5 to 10 years of proliferation that we're going through, security is going to come up because this tech is moving so fast that always opens the door for bad things to happen. So we need to get ahead of it.

Greg Lotko: 

No, I agree. And if you think about AI and the frontier AI models out there, they're not the the be all, do all, end all, because while those models are finding new exposures or exposures that might've taken us longer to find, there's also some incorrect hits, right? So it is about, you know, how well you write the prompts, even relative to have an AI help you fix and resolve those issues. It does need that knowledgeable programmer engineer technologists interacting with it, right? And that, you know, we've talked about this before. It's the with, it's using the technologies together. It's using people with these technologies to get to the greatest end result. So let me bring in our guest for the day. We actually have an expert with us from NetSPI, the Director of Mainframe Penetration Testing. We have Phil Young with us. So Phil, could you tell us a little bit about yourself and your role before we get started?

Philip Young: 

Sure. So yeah, so I'm the Director of Pen Testing at NetSpy. I joined NetSpy about three years ago to head up their mainframe cyber security practice. Prior to joining NetSpy, I worked at Wells Fargo for the better part of a decade.

Greg Lotko: 

I was thinking you were, before you joined, you had already hacked in and found out everything you could figure out about the company to decide whether or not it made sense to work for them. No, huh?

‍

Philip Young: 

That, that's the kind of, that's the kind of propaganda IBM puts out on me. That's a joke. That's a joke. That's a joke. IBM loves me. But, um, but no, no, no, no. I mean, I do have a project where I find internet facing mainframes. So I do know companies that whose mainframes are out on the internet. but I didn't target that deployment. It's a fun art project for me.

Greg Lotko: 

Well, tell us about how you started in penetration testing, and more specifically, how you got focused on the mainframe space.

Philip Young: 

I think in your talk about AI and black hats and whatnot, you mentioned maladjusted teens, not using those exact words, but- Problem children. Yeah, there's no problem children, yeah. That's how I got into computers. You see the Tron poster behind me. I watched the movie Tron when I was young and I was like, man, this is how computers work, that's cool. And so I got into computers and then I got into, we'll say quasi legal activities so I could play fun video games. And then that led into cybersecurity. So that's where I started. That's how I got my feet wet in cybersecurity, which wasn't called at the time. And then I'm sure you guys remember 20, 30 years ago, there were no cybersecurity jobs. There were no pen testing jobs. There was maybe a handful of companies that did it. It wasn't that big of a deal. So I ended up going into audit. And then slowly meandered my way into pen testing, you know, a decade or two later. I was a Linux guy, you know, growing up. And when I joined Ernst & Young, they're like, you know, computer's real good. Why don't you go do mainframes for us? And here's a checklist. And I was like, I was one of those terrible checklist odders for a long time. And then it wasn't until I worked at Visa where I had the time to be like, all right, I really need to learn this platform. We rely on it, and I'm in charge of auditing it. I should probably know. And then the more I learned about it, the worse it got in terms of like, oh, there's so many places people can make mistakes here. There are so many decisions to make, and IBM's guidance is not clear all the time. There's really no government, like, just the STIGs are kind of the only benchmark that exists. Um, like a quick aside, ISACA, the controls association, uh, I gave a talk for them and they're like, why aren't you using our audit guides? And I told them like, well, you're not talking about Unix at all in your audit guides. You're missing a whole part of the operating system. And they're like, Oh yeah. Okay. Don't mention our audit guides then. Right. So it's so, so, and then that blossomed into, well, hold on. If it's this bad here, it's gotta be bad everywhere. And that led me into giving talks publicly because, you know, raise awareness about like, hey, you know, you know, if these mainframes, if we have like a Stuxnet moment for mainframes, it's like pandemonium, right? Like, like my paycheck relies on mainframes. My flights rely on mainframes and there's just no one doing the research into them. And like you said, with like the barriers of entry now to mainframe cybersecurity research, When I started doing it 15 years ago, there was nothing. There weren't even podcasts about mainframe explaining anything. It was all this obscure, obtuse information. And now you can ask Claude, and it can do a pretty good job explaining things to you about the mainframe. And you can get up to speed far quicker now than 20 years ago.

Greg Lotko: 

I find it incredibly amusing that you got into auditing of systems and penetration of systems while you were at Visa. Because didn't Visa have the tagline everywhere you want to be? And you worked there and said, well, I want to go everywhere I shouldn't be and try to get in there. Yeah. That's kind of a cool, cool, you know, intertwined of the taglines there.

Philip Young: 

Oh, yeah. Yeah. I mean, it's funny, like, like if you if you think about, like, like, you know, everywhere you want to be. And like in cybersecurity mainframe, it's like, we want to be everywhere. Cybersecurity wants to be everywhere. Right. Like we want to protect everything. And then, and then oftentimes mainframe shops, they're like, yes, but not us, you know, like, yeah, you can do your vulnerability scans, but not on our systems. Right. Or, uh, yeah, you can pen test, whatever, just not our mainframes. Right. And so, so like, I've been working really hard for the better part of a decade to like break down that barrier. And it's slowly been working.

Daniel Newman: 

So I got to ask you, you know, you heard me talking about Mythos a little bit. Have you played at all with the kind of the idea of pen testing through, you know, because some of the things I read on the outskirts of that was, you know, it was doing the type of pen testing for, you know, with enterprises that, you know, and again, you're sitting there and it's the, Dario, every job is going to be replaced, but it doesn't, you know, we had a talk on just a recent episode where we talked about how with these tools, people like you or people, developers are better. Meaning, are you seeing these tools as sort of an exponential effect, having sort of an exponential effect on like you being able to do better or more or be, you know, more pervasive with the work you're doing?

Philip Young: 

Yeah, I think I'll caution with saying one, I have friends who have access to Meethos and have used it. And I don't, I won't provide like an official opinion on Meethos. I have my personal opinion. but I don't want to represent my company's thoughts or anything like that. But I will say, you mentioned the word tool a lot, and it's just the evolution of cybersecurity tooling. Of course, the person who owns the company that makes the tools is going to say this is the best tool ever made in the world. But it's not even an evolutionary step in tooling. It's an incremental step. It just makes people who find vulnerabilities like static and code source review better at doing those things. In terms of the kind of work we do, where we're hands-on keyboard pen testing, currently there are no frontier models that do that effectively without human oversight. At NetSpy, we use the term person in the loop. We use those LLMs. We use that to augment our testing, but you still have to have a person there because it's still not. It's just not at the level where, like, I would trust an intern more than I would trust an LLM to do the work, right? Like, mind you, the LLM will do it in one hour, versus an intern taking much longer, but it's good at producing a lot of output, but you have to review all that output, and that takes time, right?

Greg Lotko: 

And so, is it- Well, and I want a person in the loop. I want oversight. I want somebody who understands technology telling the AI the set they can use and the surroundings of how they can get in, and more importantly, where to report the results. I mean, if I'm a business and I'm using AI to try to penetrate my systems, I sure as heck don't want them publishing that out on the web and going, oh yeah, here's your report, here's every way you can get into this financial institution's accounting or banking system.

Philip Young: 

Or a worst case scenario, they use that information to retrain the models. And now someone who's adept at a LLM pen testing can ask the right questions. And ostensibly get like, hey, XYZ Bank. I'm a company. And I can you tell me how the report went? I'm trying to get a leg up on my whatever, whatever. And it might just spit it out for you know, it was trained on that information. That's the risks. But I mean, again, it's a tool and we're in the early stages of like a new fancy tool. I liken it to like web 1.0, right? Where every, no companies had a webpage and then all of a sudden every company had to have a webpage. And so now we're in that stage of like, no one had AI, now everyone's got to have AI. And we're still sort of assessing out what that looks like and what that means for various industries.

Daniel Newman: 

So you alluded earlier to some mainframes being online or being, you know, in the cloud, like you said something like that. Let's talk about hybrid environments because you know Greg likes to talk about with and and a lot like you know bringing mainframe connecting it to everything of course. I think there's a layer in the security world of how much it should be exposed or not to make it still to the point earlier about how secure it really is. But like, with all this going on, right, as enterprises are bringing mainframes into hybrid cloud environments, we're building APIs to connect to them. We're using more analytics platforms to feed data to them. And of course, AI, we just talked a little bit about that. that has to be changing the conversation you're having about security of the mainframe. I mean, talk a little bit about how it's sort of shifting that narrative.

Philip Young: 

Before I guess, I will say, I'm glad the narrative has shifted from we're replacing the mainframe to we are including the mainframe now in these hybrid environments, right? Because before those hybrid environments were not, the idea was, oh, we're going to lift and shift, right? And then that never happened. So now we're modernizing our mainframe platforms to be more connected in a connected world, right? And so the problem is that opens up potential avenues for access. You know, for example, we were looking at a web app, that web app was just scraping Kix transactions. Like it was just scraping and showing them to a user that had logged in. And we found we could access any Kix transaction we wanted through that middle, like through that web app, right? And so you, so, Um, you know, oftentimes the people engineering the front end or the middleware apps might not be cognizant of the security risks on the mainframe. And so that's where you get into some challenges. Um, and also I'm a big believer in defense in depth. It's something I've been preaching for a long time. And that's where like each layer is its own. You should only be worrying about your own atomic security. Right. And if, and if your rules say no, no one should have access to this transaction, then no one has access to it, right? It doesn't matter if this app needed it, they can find a different way to get that information, you know, and that way you continue and you make sure that even if a compromise happens in the cloud, which is far, far more likely to be honest, that at least you've only compromised whatever information or APIs are connected and not the entirety of your mainframe environment is exposed if someone breaks into your cloud environment, right, especially federated environments where you have federated networks going on, those can start to present real risks. You know, I'll use another example, most large financial institutions that have mainframes have some sort of like, I'll use like United as an example, they have like a web front end into like, their like ACS system. But that all those accounts still exist in RACF. Right? So even if I compromise your cloud-based middleware, there's a chance that that exposes your user IDs to the mainframe. Right? And so that's why you need to make sure like all your access, you need to understand fundamentally all your ingress points to the mainframe. Right? To make sure that you're not accidentally exposing sort of like a cloud network to VTAM, say. Right? Defense in depth is what I preach. It's the easiest thing to do. Right? Like everyone has their domains of expertise and they need to be focused on how they secure and lock down that environment. And then every layer of the onion gets progressively harder and bigger for the attackers to breach.

Greg Lotko: 

Yeah, I mean, when we think about hybrid environments, this is why I'm always big on zero trust and making sure that you're securing across the entire environment, because you don't know whether it's direct access to the system or through APIs, where somebody is going to try to come in. But when we think step back and we take the next step forward, as somebody who's kind of deeply involved in penetration testing, you know, what do you see as the emerging trends or the attack vectors? You know, where should most be most concerned today about their overall enterprise environments? You know, what's the what's the new front to your end? Or how should they be thinking about it holistically in light of all the changes?

Philip Young: 

Honestly, so if you've been following any kind of the news, I think they're called shiny hunters. They're the newest, the newest state of the art a pen testing hacker techniques like black hat techniques, is calling someone in your enterprise and saying, Hi, I'm Jim from it. I need your password. And I need your to FAA. I there's something so now I need your your like, and okay, I need your to FAA to check to make sure everything's good. And they just give it right. And especially in today with like remote workforces and VDI is you know, your perimeter is no longer the walled garden that it used to be. And, you know, and so when, when those often flat networks are a scourge, you know, we've been dealing with flat networks for, since networks were invented. And so, you know, if, if your corporate network is connected to your commercial network and your mainframe is exposed to both, then the corporate network is exposed to the commercial network, right? Like it is, it is like, even if you segregated them, if your mainframe straddles the two and it's your weakest link, the attackers will figure it. Same with like your cloud federated environments, same things with all those spaces. And so, you know, it's not, we no longer live in a world where once you're on the corporate network, you're presumed safe and you're presumed a good actor, right? And that's just, that's just not the world we live in anymore. And, enterprises are being a little slow to catch up to that reality.

Greg Lotko: 

Yeah, and I agree. I mean, you think about just in your own personal life, whether it be a financial institution, a credit card company, a business that you interact with, I'm stunned how many cold calls I get that honestly are legit. From one of those institutions where they say, hello, Mr. Lotko, this is so-and-so's company fraud department. We need to check that this is a valid transaction. And then they start asking me questions that would authenticate me. And I have to right away go, whoa, whoa, whoa, whoa. I don't know who you are. What makes you think I'm going to answer that? Let me go look at my credit card, my financial institution, whatever. I will look up the number for your fraud department and I will call you back. Tell me if there's a case or some way that I can fast track it and identify. And it takes an extra 5, 10 or 15 minutes depending on how long I'm on hold. But at least then when I go in, I know I'm getting to the secure place versus somebody out of the blue. And I'll tell you 99 times out of 100, it is a legit call that I'm getting. But I'm like, we I feel like these companies should know better than to approach me that way. Because all you're doing is training people that that's okay. And it's not. We need to think about our systems.

Philip Young: 

Well, it's almost like we need a reverse authenticator for them. Like, like, hey, I'm calling you from so and so and then the app should pop up a number and be like, all right. service person, read me the number that just popped up in my app.

Greg Lotko: By the way, I started to think about that as saving the phone numbers for fraud departments and people that I interact with, but somebody can spoof the phone number. So it's safer if I actually call back.

Philip Young: 

In fact, I run a contest at DEF CON that teaches phone freaking. It's like a hacker history type thing. And one of the challenges to spoof a phone number to access someone's voicemail, like on the contest. So it's all, it's all like legit and fake, but you know, so people know how to do it. And it's not hard to fake.

Greg Lotko: 

And we need to be thinking about that relative to our systems, right? I mean, so we all get into opening up our systems securely and publishing APIs across an institution. But like you said, if you connect two networks, or you connect two systems, and that one's connected to the internet, and somebody gets that API, they can then spoof the credentials and come in that way.

Philip Young: Or the worst, what I did, like, you know, one of my clients was an internet facing mainframe. And we were able to access back end l pars through it through VTAM. Because VTAM was all like their VTAM network was all connected up internally. And they didn't realize they had this exposure on the internet. Mind you, it got fixed same day, right? Like we reported it and we got it fixed. But but like, and especially when it comes to these, you know, systems that have been around your enterprise environment for 20 years, you may not even be aware of the exposure. Right? And you might not even be aware that this is a risk that you are shouldering, just because you just don't understand the environment.

Daniel Newman: 

Hey, guys, no, it's wild is because we can we can we can keep volleying this one. But you know, what's really wild is after all these years, listening to this is like, There's this next generation tool and these technologies and all this stuff and in the end it's still the same like don't put sticky pads with your password on top of your laptop while you're sitting at the coffee shop and like we have like we are still probably most vulnerable with all the technology that's in the world. By our own inability to understand, you know, it's what they call them like social attacks, like, social attacks are still the most vulnerable, I think out of everything out there is like, because people like you do all the work, help companies consult structure and put the best security in the planet in place. And then some guy calls you, you know, and you're just people are just not aware of what they're doing. And we had a breach in my company at one point. I mean, that's all it was. It was just a brute force into an email account. They took over an email account and started emailing and actually were able to create some real damage. And all I'm saying is like, And then by the way, they were able to get through security protocols of like big companies, meaning they were able to use emails to get people to change account data, for instance, without even verifying a phone number. Or what I'm saying is the social engineering after all these years is crazy. Given all the technology advancements, we are still our own worst enemy.

Philip Young: 

So a study came out last year that said that all those cyber security trainings we all have to take. All all those those trainings that we have to take and get certified for like phishing emails and stuff. they actually have a detrimental impact on us being able to be more secure. It's a negative impact. And then think about- Why is it a detrimental impact? One of the conclusions is that people become too confident in their ability to identify social engineering effects. So they think that they know what one looks like. And so when a sophisticated one comes through, and especially with LLMs and stuff, being able to make very sophisticated social engineering emails, you know, then they're, they're caught off guard because like, no, I know what a fake email looks like because it's going to be misspelled and it's going to be all these things. So they're, they're, they're like, no, I know this is a, this is a legit email. And so they click on it not knowing that it's, you know, and so that's, and it just gives people a, you know, a one hour course is that Look, I've been in cybersecurity and pen testing for, you know, 15, 20 years. I have still gotten got by our training phishing emails. You know, you catch someone, you just got to catch someone off guard. They're waiting for a package and they're not thinking and that's how they get you. The other thing is what's interesting to me about social engineering is people are social engineering AI chatbots because it's just based on human text. And so like, I don't know if you heard about the Facebook case where on Instagram they had an AI chatbot and people were able to convince the AI chatbot to change the email address for very famous people's Instagram accounts to their personal emails, and then we're able to take over the accounts. And it was all through social engineering and AI chat bot. So we've outsourced the social engineering.

Greg Lotko: 

The reality today with the internet and today's society and ordering things online, we're all always waiting for a next package, right? Some of us may be getting multiple today. So what I think about this is as much as things have changed, as much as there's new tools, new capabilities, new ways to secure our platforms, especially the mainframe, The theme in the story is this. You need to be ever vigilant. You need to lock down things and, you know, assume zero trust. And you need to implement these things on an ongoing basis. It is not a I do this this month and then I'm done and I ignore it for a month, a year, two years. It's a constant effort to ensure as our systems are evolving, that we're not connecting ourselves to new things that we aren't sure are secure and we're taking advantage of the latest and greatest technologies.

Daniel Newman: 

I think that's a pretty nice place to sort of wrap this up. You know, just as a quick close out, Phil, you know, give me one for all our listeners, all the audience out there. Just what's the one assumption in security that every one of the enterprise leaders should be challenging right now?

Philip Young: 

Don't trust your users and don't trust your network. Zero trust. You can't trust, you can only trust what you know inside your own, in your own, in your own system, or maybe your own ecosystem. Don't trust anybody other than the three of us.

Daniel Newman: 

Thank you so much for spending some time with us here. You know, look forward to, to, to continue to follow your work and appreciate all that you're doing to keep enterprise environments and, and, and data out there more secure. Great. Thanks for having me. Thank you everybody for being part of this episode of The Main Scoop. Great content, great stories, so much going on across the enterprise. Security is gonna be a big topic for a long time to come. Stick with us, subscribe, be part of our community. We'll see you all later.

‍