Agentic AI was the buzzword of the year in 2025. Everyone wants to figure out how to use agents, but how do you know how much authority to give them in your SOC?
Check out this post for the discussion that is the basis of our conversation on this week’s episode co-hosted by me, David Spark, the producer of CISO Series, and Steve Zalewski. Joining us is our sponsored guest, Cliff Crosland, co-founder and CEO, Scanner.
Got feedback? Join the conversation on LinkedIn.
Huge thanks to our sponsor, Scanner
Full Transcript
Intro
0:00.000
[David Spark] Agentic AI was the buzzword of the year 2025. Everyone wants to figure out how to use agents, but how do you know how much authority to give them in your SOC?
[Voiceover] You’re listening to Defense in Depth.
[David Spark] Welcome to Defense in Depth. My name is David Spark, producer of the CISO Series, and joining me as my co-host is Steve Zalewski. Steve, let’s hear your sign on.
[Steve Zalewski] Hello, audience.
[David Spark] If you see Steve in person, feel free to greet him that way.
[Steve Zalewski] [Laughter]
[David Spark] He has received a, “Hello, Steve,” and he appreciates it.
[Steve Zalewski] [Laughter]
[David Spark] Our sponsor for today’s episode is Scanner – log everything, detect without limits, search instantly. You’re going to learn more about just that a little bit later in the show, but we’re actually going to be talking a lot about this throughout the entire show.
So, if you’re listening and you want to learn about dealing with logs and dealing with the SOC and dealing with agents, you’ve come to the right place. So, let’s talk about this. AI agents hold a lot of promise in cybersecurity, a lot of excitement.
There’s the idea that they can do a lot of the boring grunt work and free up analysts to do more interesting investigations, but for them to be useful, we at some point need to give them more autonomy. How do we, Steve, balance that with human oversight to get the productivity gains we’ve all been promised?
I know you asked this question at the LinkedIn community, and you got an overwhelming response.
[Steve Zalewski] Clearly, we touched a nerve on this, and what was so fascinating from my perspective was it wasn’t whether we need to do it, it’s like we’re past that. And so, we had 30,000 views of that, that generally were then talking about the “So what, now what?” Really interesting.
[David Spark] Well, to join us in this very conversation and thrilled that he’s here and they’ve been a spectacular sponsor of the CISO Series, it’s our sponsor guest, the co-founder and CEO of Scanner, Cliff Crosland. Cliff, thank you so much for joining us.
[Cliff Crosland] Thank you, I’m so excited to chat about this.
How would you handle this situation?
2:09.386
[David Spark] Andrew Wilder, CSO of Vetcor, said, “It’s the crawl, walk, run model with the human in the loop that builds trust and adoption.” And by the way, that’s kind of a generic theme through all of this. Supro Ghose, CISO over at Graphene Security said, “AI SOC agents should follow the same kind of maturity curve we’ve used for years with SOAR and network anomaly detection.
Start read-only, tune, observe, tune again, and then only introduce controlled action. In the early stages, I’m comfortable letting agents handle low risk, high volume tasks, the things that analysts do dozens of times a day. For example, flagging VPN access from outside CONUS, checking against an approved user list and sending a notification to the manager.
These are deterministic workflows with clear guardrails. Once the agent proves it can consistently execute these repeatable micro decisions, then we can slowly raise the bar, not by giving it broad autonomy, but by expanding the library of tightly scoped actions with human-in-the-loop checkpoints.
To me, the balance is simple. AI earns authority the same way junior analysts do – demonstrated reliability over time, starting with the basics and gradually moving up the stack.
And then last, Asaf Wiener of Mate Security, “As we gradually move the automation slider, we can intentionally reduce human involvement, but always in a way that’s aligned with the organization’s need, pace, and risk tolerance. While defining the line today is important, that line will inevitably shift over time, focused on principles of adoption, grounded in real needs and requirements rather than fixed boundaries.” All right, this was a lot of quotes, but man, I think they all hit it on the head, Steve.
[Steve Zalewski] Yes, and that’s what we were talking about. And what surprised me was last year, there was a lot of conversation about fighting adoption, why it’s going to cost us jobs, why we can’t trust it. And we’ve seemed to have crossed over the chasm to people simply saying, well, it actually has a value proposition.
And we’re not just worried about replacing people, we’re talking about autonomy, we’re talking about accountability and responsibility. And we’re starting to see that the value proposition is there in a way that it accommodates our ability to do a better job.
It doesn’t replace the humans in doing a better job.
[David Spark] That’s a really good point. And I also like this analogy to how we teach junior analysts. We’ve been hearing this again and again. AI should be treated like we treat our own staff. You don’t let them run wild, don’t let AI run wild, and I know you feel the same way, Cliff, yes?
[Cliff Crosland] Yes. I think a useful model for thinking about AI agents is to think of them as like the world’s smartest intern, where every session where you work with an agent, it’s their first day on the job. They know nothing about your actual business context, but they know everything about maybe security, software engineering, they know like ancient Egyptian history.
[Laughter] These LLMs know everything, but you need to treat them with a lot of care the same way that you would treat an extremely smart and eager intern who doesn’t have the full business context yet.
[Steve Zalewski] You know, here’s the crawl, walk, run analogy, Cliff, is the one I get with because it’s more like a child to a teenager to an adult, I think is a better analogy because we look at them as idiot savants when they’re three months old and can read, and we go, “Wow.
Oh, my God, it’s amazing.” But we realize the limits to they may be able to read, but they’re not so good talking, okay? And then we get to the teenager stage of having a whole different perspective about what the skillsets are and how to do that. And then adult, I would argue right now that what we’re seeing is AI has moved beyond the three-month-old, look at how amazing my child is, to approaching teenagers, which is we need to have them mow the lawn.
And we need to figure out the skillsets they need to mow the lawn because that’s a value proposition that we’re ready to adopt.
[Cliff Crosland] Yes, like the same way that like a teenager at a summer job, they need supervision, they need a manager, they can’t make unilateral decisions for the company or like make critical decisions that might make customers really upset, like banning an IP address on their own, giving them some more autonomy, but making sure that someone supervises and the manager is there to really make sure that any critical decision gets reviewed.
I still think that’s like essential, but I think that’s a good model. They’re teenagers now and they’re not adults, so we have to be careful.
What is the most critical issue?
7:00.093
[David Spark] Betha Aris Susanto of DTrust said, “I draw the line based on blast radius, not AI capability. AI should own analysis, enrichment, and memory. Humans keep authority over actions that affect availability, trust boundaries, or external accountability.
If a decision must be explained to an auditor, regulator, or the board, it stays human. Read-only by default is not conservative, it’s correct.” Rock Lambros of RockCyber said, “The read-only versus write-access boundary is what most teams start, but it’s the wrong line.” So, not agreeing here.
“Reversibility matters more,” I like that. “Quarantining a host is a right action I’d let an agent execute unsupervised because I can undo it in seconds. Disabling a production service account, same permission level, completely different blast radius.
So, teams aren’t really debating autonomy levels. They’re discovering they never built the authorization model to support tier trust. Most AI agents inherit the API credentials of whoever deployed them. That read-only triage agent probably has more standing privilege than any human analyst would ever be granted.
That’s the real risk, not over-automation. It’s the well-meaning super burned-out SOC engineers are deploying agents with permanent credentials some of them wouldn’t hold themselves to to reduce their cognitive overload.” All right, so this is a really interesting take here, Cliff.
It’s about blast radius. Like if this agent went rogue or made a bad decision, how messed up would we be? And I think that’s a very good line to understand, yes?
[Cliff Crosland] Yes, it’s really interesting. All of the customers that we work with who have built agents on top of our tool and on top of other tools, they’re definitely leaning toward not allowing agents to do anything with write access. And in fact, they’ll go as far as saying that in the investigation, don’t make your recommendations too strong because then you’re going to encourage the human analysts to do writes that they shouldn’t be doing.
There’s like maybe even less trust. [Laughter] Even if you have a read-only agent, don’t let that agent be too persuasive to cause the wrong kind of actions to be taken.
I think what really needs to happen before we could trust agents to do writes and to take action, maybe it’s fine to do things with little blast radius that are easily undoable, I think that probably is fine. But one thing that needs to be solved for agents is they need to remember things.
They need to learn on the job. That’s not something that’s very easy to do with agents right now. The context window gets refreshed from scratch and you have to like teach it from scratch every session. I think as new innovations come out to let them learn on the job and not just be a fresh intern every day, then I think it’ll be easier to trust them with taking more and more serious actions, but until then, we have to be careful.
[David Spark] All right, I’m going to throw this to you, Steve. Beth at the beginning said read-only by default, it’s not conservative, it’s correct. But then Rock says I’m okay with read/write just as long as it’s something I can reverse. So, essentially it doesn’t have that blast radius problem, which I think he’s still agreeing with Beth.
It’s just the write part is just nothing that can cause real damage, yes?
[Steve Zalewski] It’s not blast radius when you hear that you think damage. What Rock is saying is speed to consequence to the business. What we’re really starting to appreciate here is what is the impact to the business, and what is an appropriate escalation with the speed with which we put the right impact on the business relative to the risk to the business.
And what I really appreciate about it is whether you go we’re only going to go read-only, because that’s how I’m testing now what the business impact is because what AI agents give me is speed of insight and speed of execution. And now I’m thinking about, “Well, what can I have it do faster to thwart the attack?” versus “When do I have to take more aggressive impact?” and “Maybe I need a human.” But the fact that we’re having that conversation is fundamentally changing the way we’re thinking about how we run a SOC and how we consequence the business, and I think that is just absolutely foundational that we’re moving in that direction.
[David Spark] Cliff, it sounds like you want to add to that.
[Cliff Crosland] Yeah, I think it’s basically the consequence to the business is the most important thing to worry about. If an agent should run the decision up the chain to the manager, and in fact like it’s probably what we’re all going to become is the managers where like agents will ask us for permission or ask us whether this decision is correct.
It might make our jobs actually kind of delightful, less deeply in the weeds with data and looking at every single log line with our own eyes [Laughter] and trying to parse them. But I think that the right model is, is still allowing agents to run the decision up the chain and make sure, like if there are major decisions that are business essential, that humans have to be involved.
Agents can’t be held accountable in the same way. You can’t really fire an agent. They don’t have the same sort of self-preservation problems. So, yes, I agree with this.
[David Spark] I mean, you can delete an agent. It’s just, they won’t really care.
[Laughter]
[Cliff Crosland] Yes, exactly, exactly.
Sponsor – Scanner
12:15.055
[David Spark] Before I go on any further, let me tell you about our fantastic sponsor. It’s Cliff’s company and that’s Scanner. So, I’ll start here with a question for you. How much of your security data can you actually run detections on right now? I mean, all your security logs end up in cloud storage like AWS S3 buckets, but only maybe 10% also goes into your SIEM.
I mean, Splunk and Datadog are just too expensive for the rest of it. So, your detections only cover a slice of what’s actually happening.
Scanner changes that. It’s a security data lake that indexes your logs directly in cloud storage and runs real-time detections on all of it. No pipelines, no re-ingestion, no schema work, hundreds of detection rules on 100% of your data, not just the small subset that fits into your SIEM.
And when you need to investigate, searches come back in seconds, not hours. So, teams at Ramp, Benchling, and Lemonade use Scanner for detections, threat hunting, and incident response. And because queries come back fast enough to iterate on, most of the query volume on Scanner now comes from AI agents.
They do the log diving. Analysts, they make the decisions. It’s 100 times faster than traditional data lakes, 10 times cheaper than traditional SIEMs. It’s loved by analysts and it’s built for AI agents. And you can check it out easily at this website, Scanner.dev.
And when you go check them out, let them know you heard about them from the CISO Series.
Where does this effort fall flat?
13:52.490
[David Spark] Anton Chuvakin, who is the host of the Google Cloud Podcast, said, “DIY, do-it-yourself AI agents for the SOC is really, really high bar. A, yes, some people do it. B, yes, some of the above people succeed. And C, the number of people in item B is really tiny to answer your question directly.
Read-only alone won’t cut the effort-versus-value balance, but this is a reasonable first step.” And Erik Bloch of Illumio said, “I don’t know anyone building an AI SOC, but people are experimenting with AI and agents to do basic summaries or using it to shorten dev times for automations or detections.
The idea that there is mass adoption around this is a marketing echo chamber.”
All right, I want to lean on that very last line that Eric said because I was saying this before, and that is traditional business model is you under-promise and over-deliver. AI has completely flipped that. All of the marketing is over-promising and definitely under-delivering, and that’s why everybody’s annoyed.
But we all see there’s value here, so that’s what we all have to aim for, and I think that’s why we’re frustrated by this. Yes, Steve?
[Steve Zalewski] Yes, and I would say in a couple different ways, right? Because when we think about this as security practitioners for AI, we all can appreciate how the bad guys have weaponized AI for offense. We see how they’ve been able to use it because we pay the penalty every day – social engineering tax and everything.
Therefore, it’s an intellectual exercise that you quickly get to, then I kind of have to adopt AI for defense. I can’t do it with humans. I can’t do it the way I have been. So, it’s not that I’m doing it differently, but I need to be able to adopt the ability of people, process, and technology.
Once we kind of approach to that, and I think that’s what happened last year, now the question is where’s the low-hanging fruit in a security organization that we, as defenders, can adopt some of this stuff in a way that we can get the money for it out of our senior leadership?
Because we’re able to be able to figure out how to reestablish that defensive perimeter. That’s kind of the exercise that’s happening in the industry. And what we’ve realized is a level one analyst in the SOC turns out to be a really good test case because a lot of what they do is relatively simple process.
That’s a lot of drudgery that isn’t action oriented, it’s investigation-oriented to then put it up the chain. And we’ve done a lot now in the last two years with a lot of companies that are building the technology to be able to demonstrate that you can buy these virtual agents to come in on that, and that’s opened the door to, well, what does a level two analyst look like?
What does a level three analyst look like? And within those existing models of workbooks, what does decision making look like?
[David Spark] And I think our frustration with the marketing here, Cliff, is giving AI a bad rap, but all of us are like but it shouldn’t get a bad rap. It’s the marketing that should get a bad rap, you know? And we know there’s value here and we want to take advantage of it.
We’re just all trying to find the line, and like what we said early on, that line’s going to shift. Your take?
[Cliff Crosland] I think one of the things that is happening is that, yeah, our expectations are way too high and we’re sort of treating it the wrong way at first. For example, you wouldn’t just say, “Hey, here’s access to all of my data. Go find threats.” That’s way too broad.
And everyone’s like, “The AI is so smart. It’s PhD level at math problems. It should just be able to do everything magically and be godlike,” and it’s not.
[David Spark] But I will say for years, we’ve been pitched in all of AI, imagine a day that you would just wake up and everything was solved for you. We’ve been given this view for quite some time.
[Cliff Crosland] Yes, absolutely. I think it may come. It’ll just take a while. I think we’d need massive amounts of data centers and GPUs out our ears. [Laughter] There may be a time when it is smart enough just to say, “Handle all of the complexity of the real world.” But for now, the right way to get an AI to be helpful is to treat it the same way you would with – my favorite thing to talk about – is treat it the same way you would an intern.
Instead of, “Go be a genius and go hunt down every threat in my system,” say like, “Here’s the runbook we give to our level one analysts. This is the runbook that we give when we get this alert. Can you try this? Can you do this investigation?” And then they take that over.
And suddenly your team, instead of hunting down threats all night long at 3 a.m., the agents can take that over for you and reduce the drudgery. So, we just have to use it properly and the marketing needs to tone down. [Laughter] If you can harness it well and you have the right expectations, it can be transformative, increase your job satisfaction because you’re not hunting down stupid low priority alerts all day long.
It can be really powerful, but it’s worth knowing how to use it effectively.
[Steve Zalewski] Do we want to talk about a brutal truth now or do we want to move on?
[David Spark] Let’s go with the brutal truth because you can’t tease us like that and then leave the audience hanging, Steve. Of course we want to hear the brutal truth.
[Steve Zalewski] Okay. Because I hear what Cliff says and I go, that’s being polite. We all got patted on the head, okay? And we said, “Oh, marketing’s doing a bad job.” And I go, but let’s talk about this. What’s happened is marketing did their job and the fear of missing out by the executive teams that AI is going to be an unfair competitive advantage to sell more jeans, that hook sunk.
And they’ve now spent three years, okay, investing in AI because they told their businesses, do it. And we’re not seeing a return on investment. And now they’re mandating, you better show me that my investment’s making money because it’s my job on the line, and now I’m putting your job on the line.
And security is feeling that pressure just like everybody else.
What’s the optimal approach?
20:03.453
[David Spark] Anatoly Chikanov of Primary Venture Partners said, “You can automate evidence collection, initial triage, and generation of curated lists of fixes/recommendations. This gets you a ton of value in terms of speed/efficiency versus doing it manually by Tier 1 folks.
The line though, should be drawn around autonomy of action because then if you are over-provisioning an agent, it itself can become something that an attacker can and will weaponize. It can also lead to unintended consequences if an agent takes actions that you didn’t think about to build guardrails around.” That is definitely going to happen.
That was my comment. But Anatoly goes on to say, “For now, given our stage of AI SOC, it’s better to keep human-in-the-loop in place and offload as much tedious, less risky work to AI agents while keeping decision-making processes more aligned with the SOC team experience.
Now in a year or two, I expect this to change as more companies will figure out how to run agents that have less risk of going rogue and access solutions become better where you can granularly scope them so blast damage, if an agent goes rogue, will be limited.” I think Anatoly summed up the whole episode quite nicely here.
[Laughter] What do you think, Cliff?
[Cliff Crosland] I definitely think that a day is coming when we will absolutely trust these agents. The way that I see the evolution of the technology is, with LLMs in particular, there was a huge breakthrough with the transformer model and pre-training on all of the data in the internet, and then reinforcement learning really helped where you can give it rewards and get humans in the loop to make the models better.
Then suddenly, the LLMs hallucinated less. That’s wonderful. But I think there’s still a technological breakthrough that’s needed for the next thing, which is I want it to remember. I want it to learn and grow. I want it to evolve from like a teenager to an adult with experience.
Because right now, every time it starts over fresh, it has to go and read our documentation, like every single time. I’m going to go look at an alert, I’m fresh on the job again. It might be a little bit more technological innovation required before we really start to trust it.
Imagine an extremely seasoned senior engineer who’s been on the job for like 40, 50 years, 100 years because [Laughter] it has the combined understanding and experience of everything. That’s going to get amazing. But I think before we really will start to trust it, I think it’s going to be like Anthropic and OpenAI and Microsoft and Google and NVIDIA.
They’re going to have to figure out how to make the models remember things and learn over time with experience.
[David Spark] It seems to me, Steve, that learning about AI is a whole separate discipline in itself, and that being that it’s coming so fast, it’s not like you’re getting AI doctorates coming out of schools right now. It’s like we all have to figure this out right now.
And I think that’s where there’s a lot of concern because there’s no one person or one entity that we can rely on. Everything’s a moving target. Steve, what do you think?
[Steve Zalewski] So, AI’s been around for 30 years. What we’re seeing is as the evolution of AI goes from correlation to insight to compelling story generation to agentic workflows, what’s happening as we’re moving up the curve, so to speak, is the capability is not linear in what it can do.
It’s becoming a growth curve that is separate, that is much steeper. And now that’s what I mean, marketing is overselling what it can do because they’re just looking at the technical capability, extracting an outrageous outcome, and not giving us the time and the maturity to figure out where the middle ground is.
Now, based on that, David, and what Cliff said, here’s about the optimal approach, Steve’s kind of perspective here, is a lot of what we’ve done in the SOC and with the analysts has been trying to find needles in a whole lot of hay. And we find more and more needles, but we don’t know if they’re important needles, we’re just scanning the hay.
And there’s this thing called detection engineering as a practice now that has emerged in security that are the SOC analysts now, not just analysts, but we formalized it that detection engineering is how do we go back and engineer our ability to look at all of the contextual data that is not just security, it’s business data, and focus on the needles that are important to our business that impact the business as opposed to finding vulnerabilities and then just demonstrate we found a lot of vulnerabilities.
So, AI, when we look at detection engineering, that’s where the two are coming together now, where we’re looking at what its capabilities are and making an engineering approach to be able to make the SOC now an AI SOC around detection engineering.
[David Spark] Cliff, I’m going to let you have to have the last word here because you were nodding your head nonstop through all of that.
[Cliff Crosland] Absolutely, I 100% agree with that. I think it’s really interesting to see that instead of the typical SOC analyst job, what happens is everyone starts to level up and start to think about how can I engineer a chain of detections in a system that has the right context for my business that’s appropriate for my business.
And one of the cool things is I think where AI has taken off the most recently is in engineering, software engineering. But it’s not the case that all SOCs are going to have to learn how to write Python code. It’s amazing how quickly you can say with an agent helping you, “Here are my high-level ideas for what I want to see.
Help me craft detection rules.” Maybe this tool has this query language, this tool has that query language. That that translation from high-level plain English into detection rules and crafting together a beautifully engineered detection system is totally what we see with all of our users.
It’s really fun to see, like an alert goes off and then the agent goes and suggests a tweak to the detection, “And here’s what I would change in the query for that detection,” and just assisting people in accelerating that detection engineering is, I think, totally where things are going and AI really helps with the detection engineering process.
Excellent. Very, very good point.
Closing
26:42.233
[David Spark] All right, we’ve come to the point of the show where honestly, all these quotes were amazing, but I’m going to put you feet to the fire, Cliff, I’m going to have you answer first. Which of all these quotes was your favorite? Just summarize, let us know the person and the quote.
Which was your favorite?
[Cliff Crosland] Yeah, it was from Anton Chuvakin. It is a really high bar to build your own agent. And then starting off with read-only first and then progressing from there is a really good approach. I think indeed it is hard. It is hard to get this right.
I think the easiest thing to do is if there’s sort of like a soft judgment that you need out of a security workflow, that is a pretty good place where AI can slot in and start to investigate or like give summaries and so on. But if you need like a hard and critical decision to be made, that is scary.
I think ultimately, yes, we will need to get there because human in the loop forever is going to get overwhelming, especially as agents investigate every single alert and ping you for a review on [Laughter] every single alert. That’s not going to be sustainable.
I think starting read-only is reasonable and over time, we’re going to progress to taking critical actions. It is not an easy thing to do and it’s going to be fun to see how people progress in building these agents.
[David Spark] Excellent. All right. Your favorite quote and why, Steve?
[Steve Zalewski] There were a lot. And we talked about so many things here about the practicality of how we’re getting there. But I’m actually going to go back to Andrew Wilder. It’s the crawl, walk, run model with human in the loop that builds trust in adoption.
And the reason why I went with that one is almost all of the conversation was simply how we are now going on that journey and where each of us individually is on it, but we’re all on it. And that is the key thing I think we took away from the discussions online on LinkedIn and here is that you’ll keep a human in the loop.
It may be in a very different way. And we’re basically at the child going to adolescence as to how we’re figuring out what that crawl, walk, run, and what human oversight is looking like as the AI agents are picking up more on capability and responsibility.
[David Spark] That brings us to the tail end of this very show. I want to thank your company, Cliff. That would be Scanner. Remember, their whole philosophy is log everything, detect without limits, search instantly, scanner.dev. Easy to find, super easy.
I’ll let you have the very last word, Cliff. Any last words about Scanner, about our topic, and are you hiring over there?
[Cliff Crosland] Yes, we are hiring. It’s really exciting times for us. It is, like you mentioned, agents are now the biggest consumers of the product. They query way more than people do. They’re so much more curious. [Laughter] They dive into everything.
So, it’s really fun for us to be building this data platform that gives extremely fast performance over huge amounts of data because the agents want to see it all. They want to check here. They want to check there. They want to check every log source and see what the full impact is of a particular breach.
So, it’s a really exciting time, and we’re hiring for a couple of fun roles on the marketing side, head of demand gen, head of product marketing. We’re spreading the word. It’s a blast to see all of the cool things that people are building on top of the product and we’re excited about all the agentic stuff that’s coming out this year.
We want to help everyone crawl, then walk, then run. [Laughter]
[David Spark] That’s awesome. That’s great to hear. All right, well, thank you very much, Cliff. Thank you very much, Steve. And thank you to our audience. We greatly appreciate your contributions, your listening, and then contributing again. In fact, this discussion was so darn good that I suggest you click on the link in the post for this episode and check out more of the conversations.
Because we only pulled about 7 or 8 quotes, but man, there was well over 100 and there’s tons of great stuff in there. Go check it out. Thank you for your contributions and listening to Defense in Depth.
[Voiceover] We’ve reached the end of Defense in Depth. Make sure to subscribe so you don’t miss yet another hot topic in cybersecurity. This show thrives on your contributions. Please write a review, leave a comment on LinkedIn or on our site CISOseries.com where you’ll also see plenty of ways to participate, including recording a question or a comment for the show.
If you’re interested in sponsoring the podcast, contact David Spark directly at David@CISOseries.com. Thank you for listening to Defense in Depth.