October 13, 2026
Who is this episode for?
Federal technology, cybersecurity, and security leaders managing human, AI, and data-loss risks.
What will you learn?
How visibility, behavioral risk, AI governance, and data protection can help reduce insider and AI-related security risks.
Looking back over decades of defending against cyberattacks, we see initial attempts at moats and protective walls being demolished by malicious actors. The next trend is the promise of Zero Trust. One could assume that AI-driven attacks will be able to overcome this verification system.
Today, we look at a new approach to protecting federal data: risk reframing. In our discussion with Beth Miller from Mimecast, she highlights the importance of visibility, including broad coverage, and establishing behavioral baselines.
One mistake Miller sees is organizations diving into adherence and detection before making sure they have a complete view of the system. This includes basic inventory before considering OMB’s mandating governance.
Because identities are easy to steal, even a strong Zero Trust environment can still pose risks. Miller argues that understanding behavior patterns is essential to protecting data. Once a baseline is established, behavioral signals can flag suspicious activity.
Instead of being dazzled by agents, Miller suggests a more sober approach. Apply identity and access to agents as you would to humans. Today, we see agents that are not systematically governed.
Miller recommends a newer response approach. Risk responses may include transferring, exploiting, accepting, avoiding, or mitigating risk rather than defaulting only to mitigation.
Listen to the podcast to better understand how federal agencies can manage human risk, shadow AI, behavioral drift, and data loss through better visibility.
In a world of AI and uncertainty today, a new look at human risk. Hit the music, Manny. Welcome to the Federal Tech Podcast, the podcast that connects you to federal technology leaders. My name is John Gilroy. I will be your moderator in the virtual studio today. Beth Miller, Global Field CISO for Mimecast. M I M E C A S T. Beth, how are you?
Speaker 1 0:36
Wonderful, John. Thanks so much for having me.
Speaker 2 0:38
Hey, the world belongs to AI, not humans. What are these humans doing here? We don’t want no stinking humans. I mean, this seems to be the. I was at Billington and spent three days there, and I tell you, people going back and forth and back and forth. I heard the word trust. I think every the word was trust, and then in many just little between the lines, it was maybe there’s got to be maybe a human involved. What they call it, in the loop or on the loop, or something to do with humans, and so let’s just dive into that today in the federal government. So, so, so, why worry about risk in the federal government, and what’s your role?
Speaker 1 1:11
Yeah, so great questions. First of all, that was a wonderful conference. I was there on a panel around AI governance, so I think we can kind of pull some of those threads that we about
Speaker 1 1:19
specifically, and I think that ties to your question, which is, you know, what is my role? So, my role at Mimecast is to ensure that we at Mimecast, I work within our security, I work directly for our CISO, and we run Mimecast at Mimecast. So we use our products, we use other products, and we talk through again all the business use cases for how to address the intersection of human AI and data loss within organizations, ours and our customers as well.
Speaker 2 1:51
Data loss. Well, that’s getting some people in the Washington D.C. area interested in that, and so, so, so Mimecast. So, tell me about this new way to evaluate human risk and how well it applied to a typical civilian agency, let’s see.
Speaker 1 2:02
Yeah, well, I mean, I I think let’s talk about the positives of what AI is doing because there’s there’s lots of ways in which we can talk about the negatives, which which I think it’s it’s appropriate to have the balance. But let’s focus on the positives. I think what AI has done for our security team and what I’m seeing with our customers as well, is really allow for that conversation to go back to the basics. So the fundamental practices around securing an organization has not changed. At Mimecast, we talk about it in in four different pillars. We say that workers, so whether it’s humans or agents have access and identities to systems and services with data and models, and so underneath that then becomes objectives and goals that you can use to measure how well you are doing. Through those measurements, you can identify vulnerabilities. Through those vulnerabilities, you can decide what you want to do with it as a business to address those gaps. So this is a you know I am mostly focused on private sector conversations, but these basics apply regardless of where you are within the government or private sector.
Speaker 2 3:14
Good, I I’ve done a lot of work with operational technology and sensors, and people have a hard time figuring out where all these OT things are. I think in the federal government they have a hard time figuring out where their IT systems are, and with all this proliferation of of shadow IT and shadow AI, so so how can if an agency wants to work for compliance with a certain area, they’re not going to be able to sufficiently know what their systems are doing unless they they have some kind of observability here, huh? So so you got maybe they’re maybe they’re being compliant with partial evidence, huh?
Speaker 1 3:52
Yeah, I mean I I think the key here is that you can’t govern what you can’t see, right? So visibility is the first step in order to To build a mature program that can help identify and protect against human and AI risk, you know, acceleration towards a negative event, and also secure you know data protection programs within an organization because they’re not mutually exclusive here. But visibility is for sure the first step in what we’re doing, if I may. So after visibility comes adherence. So once you can see everything, then you have to make a decision around what’s acceptable or unacceptable in your environment. So that’s adherence. And then once you’ve decided what’s acceptable or unacceptable in your environment, that’s then detection. So you have to be able to detect what’s unacceptable in your environment. So this goes back to your, you know, exact point there around this movement from shadow IT, which we’ve done before in the past. It’s very challenging, but it’s possible to wrap your head around now to shadow AI. So getting the fundamentals really straight around visibility, which you can see, adhere. What is it? Acceptable and unacceptable, and then what can you do about it from a detection standpoint? You know, you can add detection slash response in there too. I’m just kind of flipping them together.
Speaker 2 5:11
Beth, I’ve done tons of interviews, and I’ve used phrases like land and expand, and move horizontally, and north, south, east, west, all kinds of phrases about this. And at that Billington conference, I went to so many sessions. They were wonderful sessions, and and and people say, well, you know, there there seems to be a certain limit to to firewalls and restrictions and security, and and people admit that the malicious actors inside the system. So one key is to watch their behavior. Okay, behavior seems to be what’s normal. It’s abnormal. So is it that is that simple? I mean, Beth Miller never orders a couch from Seattle, and so for credit cards being used. So so what is your? I know you’re written a book about this. So what is your perspective on that, and how can my federal people benefit from that view?
Speaker 1 5:58
Yeah. So absolutely. I think when you’re talking about behavior and behavioral drift, you have to be very concise in understanding the granular level around what normal looks like. So there are comparisons. Banks are very good at this. Remember when you used to use your credit card at a certain store and it would contact you immediately and say, “Is this you? Are you making a purchase? I mean, around the holidays, I would get all of these different notifications, but those have since gone away, and they don’t really announce themselves. It’s just that the algorithms have kind of caught up to what the baseline normal looks like, right? So we’re talking about again the fundamentals of making sure that you understand what good looks like in your organization from individual to individual, compared to themselves over time, individual to their cohorts compared to themselves over time, and then individuals compared to like general population over time, and that gives you enough of a you know statistical probability to be able to make some predictive analysis that then you can do something about it when the behavior drifts outside of those expectations. So again, that that’s kind of one one way to look at it. Then I think that there’s also this component. You know, if you want to do something about it, being able to decide what of what you want to do about it can be automated in a meaningful way. So you know, if there’s behavior that you’re seeing over time that is drifting. At what point do you want to start coaching or nudging or auditing? Did you really mean to do this? If so, you know, is there another way to do this so that you’re you’re starting to identify those workarounds that employees are doing, maybe rightfully so to get their job done fast, but that are putting your company more at risk. And so, layering on education, updating policies, both administrative and technical, are really key before you just add on, you know, buying buying new tools to solve the problem. So
Speaker 2 7:49
Beth, you talked about pillars earlier, and you talked about workers and humans and agents. And if humors exhibit behavioral patterns that don’t comply with normal, I would imagine that agents could do the same thing, couldn’t they?
Speaker 1 8:02
100% So if we get, if we go back and we establish the fundamentals, and we say that we’ve been doing everything right when it comes to an insider risk program, right? And an insider risk program is looking at again behavior over time, establishing acceptable norms, and then and then creating different off ramps for companies to do something proactively or reactively in in certain cases as well. Then what we’re what becomes significant is that that work that you’ve been doing can now be leveraged to tether the AI agent’s behavior back to the human. So if you understand who your riskiest humans are within an organization, then you can correlate what AI they’re using, and then you can you can segment out that AI as being maybe a higher level of risk or lower level of risk depending on that correlation. So I think it’s an extension of that human behavior. It’s always going to go back to that human and the judgment that’s encoded through the provisioning decisions that take place,
Speaker 2 9:03
how do you prioritize who or which agent to watch, which one not? I mean, is there? I mean, how do you do that anyway?
Speaker 1 9:11
Yeah, I mean, I’m going to take a step back and say, well, it depends, and I and I hate leaving you know the audience with an it depends answer, but it depends on the foundations that you’ve established within your organization, what we recommend as best practices is making sure that you understand the use cases behind it. Right. So, what are the use cases in your organization that represent the highest level of risk, and how do you find alignment between your human and your AI risk management programs to make sure that you’re you’re creating these tier systems that you’re prioritizing the right use cases for what your agency cares about the most. The second thing is making sure that you understand what risk looks like from your stakeholders’ perspectives. So security is notorious for for for kind of siloing themselves and then kind of mitigating problems before they. Maybe surface up to the business, and I think some of that behavior is counterproductive in this environment because security can only do so much about it. It becomes more of a stakeholder engagement discussion and that business alignment. Right? I talk about it. This is an acronym I use in our at our organization, but also with different customers that I consult with, the acronym is called Team. So it’s a reminder that it takes teamwork in order to to do this well. And by team, I talk about as T E A A M. So you can transfer risk, you can exploit risk, you can accept, you can avoid, and you can mitigate. A lot of security professionals tend to go straight to mitigation and think of risk management as mitigation as being one and the same. When actually it might be a strategy you need to transfer it to another stakeholder, you need to accept a certain level of it, you need to avoid it. So thinking about it more a balanced approach is what I think will continue to allow for the endurance level of the security team to to maintain you know the the levels that they need in order to meet whatever the future challenges are going to be.
Speaker 2 11:17
I do a lot of work with microphones. I think about voice all the time, and there are ways now that someone can take Beth Miller’s voice, capture it, and produce reasonable quality that would make people think it’s actually Beth. And what I’ve heard is that now banks are being scammed by voices, people calling in saying, “Hey, this is Beth Miller. Send that John Gilroy $20,000. I mean, there’s there’s so many attacks. There’s so many new ways to do it. I mean, two years ago I wouldn’t have dreamed of doing that, but they’re doing that now. So, so from a federal agency perspective, is is training the answer? Is is it reinforced every six months training? Yeah. I mean, I mean, it looks like compliance with standards and guidelines is pretty good, but once the malicious actors inside the network-all that goes throughout the window. I mean, CEO sends a voicemail saying, “Beth, yeah, you can do it, and and bad things happen. I mean, how do you train people to avoid that?
Speaker 1 12:10
Oh my gosh, what a great question, John. So I think yes, there is training, there is awareness, there is in the moment, you know, coaching that needs to take place as well. But you can’t possibly plan for all the different scenarios. What you can do is you can revisit some of those policies that you have. Some of those, let’s just call it, let’s just use incident response plans as an example. Pick on that for a second. A lot of times when I go and consult with organizations, you know, you ask and say, okay, if this event occurs, what is your strategy to to help from a risk management perspective, you know, walk me through the different steps. And they have this beautifully. I mean, now it’s it’s. I talk about it as if it’s it’s actually 100 pages you can see. It’s you know now it’s just a file that you reference. But once you start asking follow on questions that say, “What do you okay? So that’s your policy, but what do you actually do? And they’re like, “Oh, we don’t actually do that. We actually do this. And so, what you’re reminded of constantly is the gap between what you say you’re going to do and then what you are actually going to do. That’s where the risk lives. So, I think John, to your point, it’s continuously educating, but also continuously communicating what good looks like, and going back to the fact that there should be very few exceptions to to what you’re trying to do from a security foundational posture management positioning, and if the exceptions do surface, there should be immediate protocols for how to consider what exceptions, how exceptions get handled, who has to see those, etc. So I think it again, as the attack surface widens, so should the accountability and the shareholder responsibility as well, so that you have those layers of defense from both an administrative control side and from a technical control side.
Speaker 2 13:54
Beth, we began this discussion with two four-letter words: data loss, D-A-T-A-L-O-S-S, and tell you that people live and die by that inside the Beltway here in Washington D.C. So you have a tool that can help prevent data exfiltration as well as manage the human risk. So tell me about this tool and what’s the name of it?
Speaker 1 14:16
Yeah, so the tool is called Insider. It it was acquired from Code 42 by Mimecast about two years ago. I was came as part of the acquisition. So Insider looks at data loss different from a lot of the tools we kind of talk about as being a DLP 2.0 tool, but it’s it’s more than that. Specifically, what we do is instead of geofencing the data, if you’ve already done the work to classify, that’s excellent. You get credit for it, but this is really focusing more on the individuals who have access to it and looking at it as geofencing the individuals. So if if you take the different use cases that we were tying back, you know, at the beginning of this conversation, making sure that you have alignment with what the. Cares about, we can take what a business life cycle looks like, compare it to what an ideal employment life cycle looks like, and then what known security life cycles look like. And so you can begin to be proactive and create common use cases across this continuum. For example, we know for sure that when an employee onboards, that is a high risk scenario where an individual is most likely capable of introducing proprietary information from their old company. Now, again, back to the importance of training, right, John? It’s really important that everyone understands good data hygiene. They understand acceptable use, you know, et cetera. But not all companies are prioritizing the same way. So we understand at Mimecast and and with our customers who have Insider as well that taking employees who are new and creating what we call watch list, and what that does is it geofences employees who have a known period of risk. So let’s say their first 30 days when they’re coming on board, and they have the opportunity to understand what good looks like within this organization. There’s additional guardrails put on to make sure that they are protected from themselves and they’re protecting the company. So that if they were to take, let’s say, a personal Dropbox and upload it to a professional, you know, a company Dropbox, there’d be a notification that would be blocked and say this is not acceptable. Here’s the acceptable use policy. Here is what is approved. If you have any follow-on questions, here’s a Slack channel, and you can contact you know our security officers who’d be more than happy to to explain why this is. So just like onboarding, offboarding. So when an employee chooses to leave as well, or whether there is an event within an organization where there are employees that are you know also let go for different reasons. Those are known inflection points for risk for that individual and risk for the organization. So you can then again begin to create these granular guardrails so you can do something about a particular known inflection point across the organization that introduces enhanced level of risk. So that’s kind of a different way of approaching what has traditionally been a data loss only conversation because you’re layering on the human behavior component, and now again with this advance of AI tools and the correlating between the AI agents that are being created by users. Now it forces you to add on one more layer. So you’ve got the business life cycle, the employee life cycle, the security life cycle. Now you’re going to have an AI life cycle on top of that. So you’re going to get into really good, you know, practices within an organization for for what you can do to manage risk across human and AI and data loss.
Speaker 2 17:44
Well, my open talked about new way to look at human. I mean, never thought of that, but that is the most dangerous play. I mean, when it comes to flying, and you go to airlines all the time, it’s landing and taking off. That’s the most dangerous part, isn’t it? Right. I mean, we all know that. And I guess when you enter an organization, I would think most humans would know not to bring a Dropbox into, but you don’t know.
Speaker 1 18:05
No, and I
Speaker 2 18:06
know, but you would think I would think no.
Speaker 1 18:08
Well, again, most organizations, you know, you again, common sense would would suggest. However, what we don’t necessarily articulate is that employees, you know, normally they change companies, but they don’t change industries. Yeah. So we know that there’s 8% of the employees that cause about 80% of the risk within an organization, and catching them at those moments of heightened levels of risk again might be the difference between having an event versus not having an event within an organization.
Speaker 2 18:38
Yeah.
Speaker 1 18:38
Or an agency. Yeah.
Speaker 2 18:39
There’s a podcast I listen to in town called Feds at the Edge, and Feds are at the edge. I mean, people are doing. Oh, I don’t want to get into this Pentagon stuff, but but there’s some war fighters are doing things at the edge that they never did before. Those data going back and forth, and I’m just I’m thinking about data exfiltration. All of a sudden, well, five years ago, no one worried about people doing stuff on the edge. They’re doing it now, and you got hybrid clouds, you got multi-clouds, you got AI bringing. What what data is an agent responsible for releasing or not releasing? This gets to be bring your thinking cap. It’s like oh, just lock down everything. I’m an old man. I remember, hey Beth, down the hall is the data center. Lock the door, we’ll be fine. I mean, that’s where’s the data center now? You know, when I think of it, data’s coming from everywhere. I don’t know the ocean, planes. Every time a plane back to the plane, the plane lands has terabytes of information. So it’s a, and and I think maybe focusing on the humans, and if you could say, well, generally speaking, the landing and come and and I’ve been in situations of you where individuals had to leave companies, and bad things can happen, and and if it can be structured in such a way that bad things can be controlled, you know, and. I think with human beings, that’s the best. Just control. Your age, teenage kids, try control.
Speaker 1 20:06
Yeah, I think also, and you’re implying this, but you know, with your airplane analogy, so much of it it comes down to communication, right? If you can express what the expectations are, what good looks like, what could and couldn’t go well, you know, or wrong, and then you know, yes, takeoff and landing. But then there’s the turbulent aspect of it as well, right? So there’s going to be a whole lot of gray area, and that’s where it’s going to be really, really important to have those relationships with the multidisciplinary stakeholders, both in your organization and externally, because sharing those lessons learned and making decisions really quickly-that’s going to be the the key differentiator between companies who can maintain and accelerate, you know, their their successful security programs within their organizations, and those that get burned out very quickly because those basics are going to matter more more so now than they ever did before.
Speaker 2 21:00
What did you see last week? ChatGPT 6.0 release. There’s turbulence. There’s a little turbulence there. Everyone wants the even people who have been with the agency for 10 years. Why am I going to try that? Wait a minute. Let’s be careful here. I mean, I never thought of it, but that’s the turbulence going on now. And guess what? There’s going to be a lot of turbulence coming that maybe 200 miles ago was clear and sunny. Now we have Chappy T coming up, and and people using LLMs on the edge. Well, what about that data? Who’s going to control that data? I mean, and can you control it? So it gets to be a more and more a human question, I think, because yeah, it’s
Speaker 1 21:38
and John, I will circle this back. Sorry for interrupting you, but I, you know, again, you brought this this point very clearly. I’ll circle this back to the very beginning. You know, you you can’t govern what you can’t see. So the visibility aspect of this is going to become more and more valuable. You know that that saying of like, well, I I if I just don’t look here, then it won’t be a problem. You know that ostrich approach, right? It’s just it’s not going to be acceptable anymore. You’re going to have to you’re going to have to see what your problems are, and then you’re going to have to do something about it. But it shouldn’t feel as overwhelming as it used to in the past, because again, the ability to to go through and manage risk as a team is going to be the differentiator here.
Speaker 2 22:19
One month ago, I was on a plane in Salt Lake City, flew right into Reagan, and I loaded up a bunch of podcasts and listened to them on the plane. That’s what I do. And what if people are listening to this podcast on a plane and they want to get more information? So where should they go, and what can they get?
Speaker 1 22:35
Great question. So I would direct them to mimecast.com, and I would direct them to the resources site. You can see under there blogs. We’ve got different training. We’ve got different just massive amounts of resource that’ll that’ll help you answer that question. Where do I start? And there’s multiple ways, but you can find your start here point there.
Speaker 2 22:56
Yeah, and that’s M is in Mary. M I M is in Mary E. E C A S T. People listening, this in the plane, you send me come by. You want a drink? What’s going on? There’s turbulence, and maybe they don’t spell. Sometimes the M’s are hard to listen to when you hear. Like I introduced you with Mimecast. I don’t know how many people actually thought there were 2m’s in there. So Mimecast, that’s good. And what I’ll do is I’ll put a link to that in the show notes. So when you go get your baggage, or if they lose your baggage, you can go to Mimecast and take a look at the resources. And they have a lot of stuff. And I think this is really a practical approach. I mean, people are talking about zero trust, but what about the humans? I mean, the humans seems to be the weak link in this whole deal. And if you can have somewhere to control that and reduce their risk, I’m sure they’ll be have a have a better experience, better flying experience, huh?
Speaker 1 23:47
Yeah, turn them into you know champions and force multipliers here. They don’t have to be the weakest link. I think moving forward, they’ll they’ll be so much added value. It’s just the communication and the education piece has to be better.
Speaker 2 24:00
You have been listening to the Federal Tech podcast with John Gilroy. Like to thank my guest Beth Miller, Global Fuel CISO for Mimecast. M I M E C A S T. Is that right?
Speaker 1 24:12
That’s correct. Thank you, John. It was a pleasure.
Speaker 2 24:14
Bye bye.
Transcribed by https://otter.ai
Beth Miller is a 20-year cybersecurity veteran and Field CISO at Mimecast. She is recognized for transforming human risk into a competitive advantage by integrating cybersecurity, behavioral insights, and business resilience. A former U.S. government intelligence officer, Beth has led critical missions in Iraq and developed insider threat programs for Fortune 500 companies. She is the author of Risk Reframing: How to Navigate Uncertainty with Resilience (2024), which introduces the Fold Framework—a groundbreaking approach to managing uncertainty. Beth’s people-first philosophy drives cultural change and empowers organizations to thrive in today’s digital landscape.

