This episode is focused around the growing challenges of securing federal systems against ransomware, emerging AI agents, and unpatchable legacy infrastructure. Steve Cooperman, President, SCOOP Cyber and Bob Blakey Co-Founder and Chief Product Officer, MIMIC, will discuss how known-good enforcement, virtual patching, and AI agent governance can help organizations move beyond detection-based security and prevent unauthorized activity before it executes.
[Anthony Jimenez]
Welcome back to Carahcast, the podcast from Carahsoft, a trusted government IT solutions provider. Subscribe to get the latest technology updates in the public sector. I'm Anthony Jimenez, yours from the Carahsoft production team.
On behalf of ScoopCyber, we would like to welcome you to today's podcast, focused around the growing challenges of securing federal systems against ransomware, emerging AI agents, and unpatchable legacy infrastructure. Steve Koopman, president of ScoopCyber, and Bob Blakely, co-founder and chief product officer of Mimic, will discuss how known good enforcement, virtual patching, and AI agent governance can help organizations move beyond detection-based security and prevent unauthorized activity before it executes.
[Steve Cooperman]
All right. Welcome everybody to another ScoopCyber podcast here in the Carahsoft studio in Reston, Virginia. We've got a really interesting discussion today.
I'll start out with a sort of a profound statement. If it's not authorized, it doesn't happen. And we're going to be thinking about that and going throughout that throughout our discussion today.
Again, super timely, even yesterday, the FBI announced a breach to several major federal agencies, NASA, Federal Reserve. This breach has been going on since 2018, undetected. So the focus has been on known bad.
How do we stop that? And with AI and nation states, it's getting harder and harder to detect. So we're going to talk a lot about known good, because that's the key.
You want your applications, your servers, everything else to be not touched. So with that, I'm going to introduce Bob Blakely, founder of Mimic. We're going to talk about Mimic today and what they do, and just have a lively discussion.
So I'm going to turn it over to you, Bob. Why don't you introduce yourself and Mimic, and I know you have an extensive, very impressive background.
[Bob Blakley]
Sure. Sure. Thanks, Steve, very much for having me here today.
So I'm Bob Blakely. I am the co-founder and chief product officer at Mimic. Mimic started out as a ransomware defense company about three years ago.
And we have since expanded our portfolio to include AI guardrails and virtual patching for applications which AI has identified vulnerabilities in.
[Steve Cooperman]
Yeah. So, you know, why start a whole company around that is a very extensive problem or just a feature. So I'd like to get your viewpoint on, you know, what prompted you to start this company and collect a lot of expertise and industry knowledge?
[Bob Blakley]
So I've been working on security for 50 years now, started in 1976. My primary motivation for co-founding Mimic was that security has always been oriented primarily toward defending corporations and government entities. And that's appropriate, right?
That's where the biggest economic benefit is. But in the last decade or so, there has been this trend of criminal actors and threat actors using the sort of pain and suffering of ordinary individual citizens as leverage to get companies to pay a ransom. And that's a really upsetting phenomenon.
And my co-founder and I just believe that it's not a problem that a civilized society should tolerate, right? Big criminal organizations or even nation state actors inflicting suffering on individuals in order to extract money from the, you know, from the economy.
[Steve Cooperman]
Yeah. So it sounds like the government has taken this very seriously as well. I know we briefly discussed, you know, executive order that came out, I think just today, 14420, really wanted to protect critical infrastructure, electric distribution systems, which were targeted as water systems in Minnesota.
So do you think, are you, you know, is the government behind this? Do you see, you know, trends that they're funding in and programs are coming out? We certainly do.
But I'd like to get your viewpoint on, you know, where the government, you know, stands and are they catching up to the problem?
[Bob Blakley]
Yeah. So the United States government and other governments are obviously focused on understanding where the most effective attacks in the contemporary landscape are and being able to defeat those because cyber attacks are not just a focus of crime. They're also a focus of nation state activity for political and military purposes.
So I think it's very important to governments these days to be able to keep ahead of the curve in terms of adversary tactics and techniques and to be able to mount an effective defense against those techniques.
[Steve Cooperman]
So there's been tremendous investment. You know, we talked about this has been going on since 2018. EDRs, where does MIMIC fit in into that?
You know, I've already invested. How does that MIMIC augment that and integrate?
[Bob Blakley]
Yeah. So what we do is very different from what an EDR does. And the clue is in the name, right?
So EDR is Endpoint Detection and Response. The idea of detection and response is based on the idea that the defender is going to recognize patterns of bad behavior, detect when a pattern of bad behavior is going on in the enterprise, and then respond in a timely way in order to interrupt that pattern of bad behavior. Adversaries these days, with the automation that they have at their disposal, move so fast that there really isn't time in the attack lifecycle for a detection and response loop.
The defenders need to be able to interrupt attack activity, essentially instantaneously, in order to prevent really significant damage to the assets that they're trying to protect. And that's the model on which we built MIMIC. So we don't try to detect patterns of bad behavior.
We look in advance at the patterns of known good behavior of an application. And then when anything tries to change that known good behavior, we just block it. We don't take the time to do a round trip to the SOC or to an AI model in the cloud and analyze things.
We just block the behavior and move on and let the application continue to do the mission work that it is designed to do.
[Steve Cooperman]
So this is for your most mission critical, you know, for the Department of War launch systems, weapon systems, the list goes on. This would allow them to literally lock them down. So when we're actually like we are now at war, these systems do not go down or change.
Is that sort of a use case?
[Bob Blakley]
Yeah, absolutely. I mean, so we are very good at protecting the most critical applications in an agency or in a commercial enterprise. But a lot of times people have a very sort of narrow view of what it means for an application to be critical.
So, you know, the application that allows personnel into the front door of a building in order to work on the mission is also a critical application. So we're applicable to a broad range of different applications. And we exist to ensure that those things are continuously able to do their jobs and to support the mission, even if an application is not absolutely the most critical thing that an organization supports.
[Steve Cooperman]
So not only does it protect against anybody changing, it protects against ransomware, really anything that would affect that mission critical application.
[Bob Blakley]
Yeah. And, you know, it bears remembering that all attacks, including ransomware attacks and including AI automated attacks, fundamentally function by changing the behavior of an application from its normal mission supportive behavior to some malicious pattern of behavior, which is going to corrupt data or steal data or interrupt the availability of the function which the application supports.
[Steve Cooperman]
So I know you guys work at the kernel level, which is really the lowest level of protect. Just for the listeners today, you know, that might not know exactly what that means. What is the kernel?
What is protecting at that level mean?
[Bob Blakley]
So the kernel is the core of the operating system that supports a computer and its applications. And the way information technology systems are structured generally is that at the core of the system, you have access to the full power of the operating system and all of the hardware that's associated with the system. And as you move up the software stack, elements of the stack prevent you from accessing the most powerful functions of the system, because number one, they're not necessary to get the job of the application done.
And number two, they're dangerous, right? So the kernel of the operating system can reformat the disk and erase everything that's on it. Lots of applications never need to do that.
And it would be a bad idea to do that by accident and an even worse idea to do it maliciously. So we run in the kernel because that's where the bad guys aspire to get to, because that's where they have access to the most powerful functions that they can use to steal data or destroy data. And if we are going to block attempts to exploit those capabilities, we have to be down there where the action is taking place so that we can spring into action instantly if something malicious starts to happen there.
[Steve Cooperman]
I know we were both out at Black Hat. For those that don't know that conference, I mean, there's the center of gravity for cyber, hundreds of vendors. Many of those are touting AI SOC.
Let's use AI to automate our SOC. Why do you think that's not really the right approach? And I think you mentioned earlier, why even let it get to the SOC?
Let's stop it before it even happens.
[Bob Blakley]
Well, yeah, that's right. I mean, so the motivation for automating SOC activities using artificial intelligence is because the detection and response loop is too slow if humans are operating it, because humans don't respond as fast as an automated program can. But the entire SOC model is, again, based on this idea of detecting patterns of malicious behavior and then responding to it.
And at Mimic, we think that that model is not going to continue to be effective because, in principle, there is infinitely much bad behavior. And the adversarial AI is capable of generating new, previously unseen bad behavior very fast and at very high volume. So even if your SOC is augmented with AI, there is still going to be a delay that is induced by the requirement to analyze the behavior and decide whether it's bad or not.
That's why we think a known good model is important, because there's only a finite amount of good behavior that an application exhibits. We can understand all of that in advance, and therefore, we can instantly block any behavior which is outside of that known good envelope. So we think that even in an age of AI SOCs, a known good model is still going to have a speed advantage and is going to be necessary to prevent damage from the most sophisticated automated attacks.
[Steve Cooperman]
So another hot topic in procurements and executive orders is zero trust, which is, as you know, based on identity. You know, same theme of question. Why is that probably not the best approach to look at it from an identity standpoint?
[Bob Blakley]
So I ran the Identity Management Service at Gartner for a couple of years, and I am not going to advise people to move away from zero trust or to move away from validating identity before they allow access. I think that's a good model. John Kinderbaug was right.
But not every attack can be defeated using an identity-based defense. And if you look, for example, at some of the recent incidents where companies have, for databases deleted by a malfunctioning AI agent, those really illustrate the problem. And the problem is that even a trusted administrator or even an AI agent that was installed in the enterprise on purpose and given a trusted administrative account can either malfunction or turn malicious and use a trusted identity and the approved enterprise tools to do things that are harmful.
So we think that a completely identity neutral, known good application behavior enforcement model is necessary in addition to the identity-based protection, which is already present in most agencies and in most enterprises.
[Steve Cooperman]
So before I lock down an application of known good, how do I know it's good before I lock it down?
[Bob Blakley]
So that's a really great question, Steve, and it's a hard question. The answer is if the enemy got there first, you don't know. And there really isn't a solution to that, right?
So Ken Thompson's Turing Award lecture, Reflections on Trusting Trust, demonstrated pretty conclusively that if a bad guy gets to a system first, he can create malicious persistence, which you can't detect later. So you do actually have to start to protect the system before the adversary gets there. And that's what we assume, right?
That when we commission a system, we are going to profile the known good behavior of the system, and we are going to have done that before the bad guy arrives. And if somebody tries to tell you, oh, well, even if the bad guy got there first, we can protect the system, you should probably be skeptical because it's more or less demonstrated that that's just impossible.
[Steve Cooperman]
So, you know, a customer or government agency could have 15 applications that are mission critical. How do they, do they deploy Mimic all at once or do they do this in a phased approach? What's your recommendation?
You know, and I think the audience will probably say, I got to have this on my systems tomorrow. So how would that?
[Bob Blakley]
Yeah, so Mimic installs in about 20 minutes on a system and our profiling of an application only takes a couple of days to build a complete known good profile. So you can get started relatively rapidly. And it's not more work to do that on multiple systems than it is on a single system because all of our profiling and policy creation is fully automated.
So your staff is not going to chew up time, you know, tweaking the knobs on the front panel of the system in order to get it dialed in. Where people often want to take a little time is after we've created the known good profile and created a set of policies, they often want to run it for a while in essentially a monitoring mode where our system says, if that had happened in production, I would have blocked it. Just because it's very important for the humans in the system to establish confidence that the defenses are working properly.
And not only is it important, but sometimes it is actually a regulatory requirement for certification of the system to operate in the designated environment. So for that purpose, we support really extensive forensics on what our system is doing, and we give you an ability to review what it's doing, what it would do in specified scenarios so that you can gain confidence and so that you can establish the evidentiary trail that allows you to certify the system for operation.
[Steve Cooperman]
So another hot topic that's been all over the news is hugging face compromise. Is that a good example where Mimic would have played a critical role?
[Bob Blakley]
So, yes, there are a couple of things that we do to defend against rogue AI. And there's sort of two kinds of scenarios here. One is the scenario where you install an AI agent on purpose, and either it gets an incorrect prompt or it starts to hallucinate and it does something harmful that you didn't anticipate.
For that use case, we implement AI guardrails, which use the known good behavior profile of the application to block the agent from doing anything that is outside the known good pattern of its own behavior. And that would have been helpful in the hugging face case. The other thing that we do is we are able to quickly develop virtual patches, which prevent the exploitation of newly identified vulnerabilities.
And I guess maybe hugging face deserves one additional piece of commentary, which is in the old days, meaning last year, we had a scoring system for vulnerabilities, which assigned a severity to the vulnerability based on essentially exploiting it in isolation. You know, how bad is this vulnerability if it is the only thing that a bad guy has access to in order to facilitate his attack? That isn't the world that we're living in anymore.
The new frontier large language models are extremely good at building chains of exploitation of a bunch of different vulnerabilities, which individually are low severity. But if you chain them together in exactly the right way, they find a way through the Swiss cheese model of security. And are able to produce a really high impact outcome, even though none of the individual vulnerabilities is high impact or high severity.
So we think it's super important to accelerate the patching of even old and nominally low severity vulnerabilities in order to help stop these very persistent, very inventive frontier AI models from finding a multistage attack, exploiting multiple vulnerabilities.
[Steve Cooperman]
Yeah. I'm curious about the virtual patching, how that works and, you know, a scenario where that would apply.
[Bob Blakley]
Yeah. Well, you know, hugging face is one such scenario. There's a bunch of others.
The way virtual patching basically works. People do virtual patching today using different technologies, right? So web application firewalls, WAFs, basically allow you to prevent the exploitation of certain network facing vulnerabilities by just blocking access to the API that exposes the vulnerability.
And that's the basic model, right? You don't actually fix the vulnerability in the binary code of the application itself. Instead, you find a point in the exploitation path and you block that particular point, right?
So that's what we do. We do it at the kernel. So in order to exploit a vulnerability, maybe you have to be able to write to a particular directory, or maybe you have to be able to open a particular network connection.
We automatically identify the system functions that create the opportunity for exploitation of the vulnerability. And then we block those and allow the rest of the non-good behavior of the application to proceed.
[Steve Cooperman]
So, you know, adversaries that want ransomware, a key way in is Active Directory. Reading up before this, you guys protect against Active Directory as well. So I'd like to hear more about that.
[Bob Blakley]
Yeah, I mean, so there are two different ways that an adversary thinks about Active Directory when he's trying to create a ransomware attack, right? One of them is, well, you know, Active Directory supports critical functions of the business or of the agency. And so if I disable it, those functions won't be able to proceed, right?
So people won't be able to log in. People, you know, won't be able to upgrade their applications, etc. So in order to create those kinds of effects, the adversary will attack Active Directory itself and try to take it out of operation.
We protect against that the same way that we protect against attacks on all kinds of applications by ransomware actors. But Active Directory is unique in the sense that it also supports the functions of lots of applications in the system. And in particular, it is often the source of truth for identity and access.
So it tells every other application in the system who is allowed to have access to which applications. And it also runs the authentication protocols that allows people to establish their identities. So for that reason, ransomware attackers will often go after Active Directory in order to help them compromise other applications in the system, right?
In order to get a privileged credential that will allow them to administer the database, right? Or get a privileged credential that will allow them to read everybody's email. So we also defend against attempts to subvert Active Directory and turn it into an attack tool.
[Steve Cooperman]
So not every system is patched. You know, I worked on a program once with ground stations. They're not touched, a 10-year cycle.
We have IoT now that is not patched, which is a tremendous threat factor and weigh in. What's your approach and thoughts to how we protect those systems?
[Bob Blakley]
Yeah, so that's based on our virtual patching technology, right? So when we create a virtual patch, what we mean by virtual is that you don't have to change the code of the application. You don't actually have to fix the vulnerability in the application code itself.
Instead, we put a piece of policy into our Mimic node, which blocks the exploitation of the vulnerability. And the code of the application continues to run the same way that it always ran. And it's just not possible for the adversary to use the vulnerability to damage the system.
And that has a couple of implications, but the two of the most interesting ones are that you don't have to stop the application and change its code and then restart it. And also, you don't have to reboot the operating system in order to activate the virtual patch. You just flip a switch in our console and the application is protected and the whole system can keep running.
You don't even have to take the application out of production. The other interesting implication is that if for any reason you decide that applying the virtual patch was a bad idea, if it interferes with some aspect of the business, you can just turn it off again and the system goes back to the state that it was in before. And again, you don't have to stop any applications to do that.
[Steve Cooperman]
So I think you did a great job explaining this to the mission owner, as well as CISOs that might be listening. So I'm not going to ask you again to summarize, but if you are a mission owner, yeah, if you think you're mission critical, MIMIC's for you because they're coming after all of us in every capacity, citizens, mission systems. With AI, the threat vector is exponential and spending time with you and MIMIC, looking at this from a whole different perspective, not that the other angle isn't important.
It just makes so much sense. Protect at the end of the day that nobody changes in your system at work. So for those out there, and I think there'll be a lot of folks that are interested in working with MIMIC, again, Scoop Cyber, do you guys do POVs?
Are there a testbed they can use? So how would they engage with MIMIC to kick the tires?
[Bob Blakley]
Yeah, absolutely. So first, let me maybe respond to your preamble. I think what really distinguishes MIMIC from other security vendors is we're not trying to protect a computer.
We're not trying to protect the bits of an application. What we're trying to protect is the process that the computer and the application support, which helps you achieve the mission, right? So we are really trying to protect the core business function, agency function that the application exists to support.
So we're really trying to protect the mission fundamentally. So how do you kick the tires on our system? Well, as usual, there's a whole spectrum of things that we can do.
We can give you a demo in our own environment. But it's also quick and easy for us to install in a dev environment or a test environment on your premises, profile an application, show you what it looks like to protect it, show you what it looks like to administer the application. And if you're interested in a demonstration of efficacy, we also have a test environment that we call the arena, and we can put an application of interest into that environment.
And we can run real ransomware or real AI agentic attacks against the application with and without our protection and show you the effectiveness of the protection that we offer.
[Steve Cooperman]
So I know Mimic is not just stopping here. It's going to be continuous innovation. You guys have a vision where this could go.
You know, a year from now, what is Mimic? And what's the long-term vision of where the company's going and addressing this issue?
[Bob Blakley]
Yeah, so, you know, I think we've just started on the journey of supporting the AI guardrail and virtual patching functionality. And I think that in the long term, you know, the position that we're going to be in is we're going to be in the position of really ensuring that the applications that are important to accomplishing a mission that are important to accomplishing a business function are continuously up and running and that their assets are not being damaged by either by malicious actors or by administrative errors and admin mistakes.
So I think, you know, there's a lot of road ahead. But fundamentally, what we're trying to do is we're trying to take an application that is doing a job in the business today and make sure it keeps doing that same job in the same way, no matter what happens.
[Steve Cooperman]
Well, Bob, I'd like to thank you on behalf of Kairosoft Scoop Cyber for coming in today. You know, very impressive. Mimic, you personally.
We're going to put up some information now on how to contact us. I encourage everybody just to learn more, get demos. We talked about POVs and Mimics can certainly play a critical role and quickly to protect your most mission critical systems.
So thank you, everyone. And we'll see you next time.
[Bob Blakley]
Well, thanks, Steve. And as you say, you know, if you want to kick the tires on this, if you want to learn more, you can get in touch with us directly at Mimic. But also, I'd encourage you to get in touch with Scoop Cyber and they can they can get you started on the journey of evaluating the solution and doing a POC, POV and get you on the path.
[Steve Cooperman]
Great. Take care, everyone.
[Anthony Jimenez]
Thanks for listening. And thank you to our guests, Steve Cooperman and Bob Blakely. Don't forget to like, comment and subscribe to Caricast.
And be sure to listen to our other discussions. If you'd like more information on how Scoop Cyber can assist your organization, please visit www.Carahsoft.com or email us at resellerpartnermarketing@carahsoft.com. Thanks again for listening and have a great day.