The federal government's most underrated cybersecurity tool

Summary

Safe Mode Podcast – Episode Summary

Episode Title: The federal government's most underrated cybersecurity tool
Date: April 16, 2026
Host: Greg Otto (Editor-in-Chief, Cyberscoop)
Guests: Philip George (Executive Technical Strategist, Merlin Group), Chris Townsend (Global VP, Elastic Public Sector), Derek Johnson (Cyberscoop Reporter)

---

Episode Overview

This episode of Safe Mode Podcast explores federal cyber defense in the age of AI—surfacing the tension between hype and reality surrounding new AI models, and making a compelling case that the “most underrated cybersecurity tool” for government remains: visibility and cyber hygiene. Across in-depth interviews and timely reporting, the episode examines how groundbreaking AI initiatives (Anthropic’s Mythos and OpenAI’s trusted access for cyber), quantum-prep efforts, and resilient data strategies are reshaping—not replacing—the fundamentals of federal cybersecurity.

---

Key Topics & Insights

1. The AI “Arms Race” in Cybersecurity

• Anthropic’s Project Glasswing & Mythos Model: A powerful, restricted-access AI crafted specifically for cyber defense. Only a select group of major tech companies (Microsoft, Cisco, and others) can experiment with it so far ([02:00], Derek Johnson).
• OpenAI’s Trusted Access for Cyber: Broader, though similarly specialized—OpenAI is optimizing existing models for security-specific tasks but is less exclusive, aiming for wider defender access ([09:55], Derek Johnson).
• Industry Hype & Reality Check: Both announcements have stirred fervor, with policymakers, tech execs, and government agencies vying for access. Yet, experts warn the core issues—vulnerability discovery, attacker rapidity—remain, and that defenders face a “bureaucratic lag” ([03:50], Derek Johnson).

Notable Quote:

“Attackers are more positioned to use AI quicker, more recklessly, with less bureaucracy than defenders are… Bureaucratic lag that legitimate organizations have to deal with.”
— Derek Johnson, [04:30]

---

Timestamps:

• [01:59–03:30]: AI model announcements, industry responses
• [03:50–07:36]: Analysis of expert reports on risks, hype vs. substantive threats

---

2. Visibility, Not “Hypebeast” AI, is Federal Agencies’ Strongest Defense

• Federal Hypebeast Mentality: Agencies, including Treasury, are eager to access the latest AI, fearing they’ll be left out of the technological leap ([07:36], Greg Otto).
• The Real Underrated Tool—Visibility: All experts, especially Philip George, argue the basics matter now more than ever. “You can’t protect what you don’t know” ([27:15], Philip George).
• Zero Trust & Inventory: Comprehensive asset inventories and continuous behavioral analytics (human and non-human identities alike) are foundational for both traditional and emerging threats—whether AI, PQC, or classic exploits.

Notable Quote:

“Do you know what the best way is to defend against itself? Figure out what’s on your network from square one and then the rest of the problem will start to take care of itself.”
— Interviewer, [27:02]

---

Timestamps:

• [13:06–32:11]: Philip George interview: how visibility, hygiene, identity, and foundational controls still trump shiny tools

---

3. AI Adoption & Data Governance in the Federal Context

• Data Hygiene is Crucial for Effective AI: Adoption amplifies existing risks—especially around identity, access management, and data integrity—because AI tools “inherit” problems lurking in the organization’s data lakes ([17:15–19:31], Philip George).
• Biggest AI-Driven Risks: Over-provisioning of accounts, poor logging, and data exfiltration by adversaries masquerading as legitimate users ([19:31–21:03], Philip George).
• Action Steps for CISOs: Build consensus with business units, target “low-hanging fruit” in data integrity, and focus on achievable wins that enhance mission outcomes ([21:03–22:54], Philip George).

Notable Quote:

“Cyber in and of itself is an enabling function. It does not exist to serve itself. It exists to equip and enable the business… Not through saying no, but this is the way, this is how we do it versus we can’t do it.”
— Philip George, [21:20]

---

4. Preparing for Quantum Threats: Cryptographic Modernization

• Priorities: Inventory and discover all cryptographic assets, shorten certificate lifecycles, and be proactive about migration and cryptographic agility ([23:39–26:28], Philip George).
• Link to Visibility: Effective “cryptographic posture management” is inseparable from core cyber hygiene—know what’s in your enterprise before you can protect it.

Notable Quote:

“Cyber 101 is to have a comprehensive and an accurate inventory of the assets that you have to protect.”
— Philip George, [23:39]

---

5. Harmonizing Zero Trust and AI

• Tension & Synergy: AI can make zero trust harder—exacerbating human process challenges with machine speed—but also helps its realization through automation and advanced analytics ([29:34–32:00], Philip George).
• Practical Take: The answer to AI-driven complexity may be more narrowly-scoped, purpose-built AI for cyber teams; both business and cyber must partner closely.

---

6. Elastic’s Perspective: Data, Standards & the Modern SOC

• Operationalizing Data with Agentic AI: Modern agencies must put their data to work—using open standards and facilitating tool interoperability is key, not just layering AI on top ([34:02–37:14], Chris Townsend).
• Standardization is Quietly Foundational: Programs like DHS’s CDM illustrate how continuous asset visibility and open data protocols strengthen the security baseline ([37:14–40:02], Chris Townsend).
• The AI Empowered SOC: LLMs trained with security frameworks (e.g., MITRE ATT&CK) are transforming analyst workflows, helping prioritize threats and automate remediation—making junior analysts more effective ([40:28–42:16], Chris Townsend).
• Change & Open-mindedness Needed: Getting the federal workforce to embrace new platforms is essential for AI-powered defense ([42:41–43:40], Chris Townsend).

Notable Quotes:

“Security is a data problem… It’s really about bringing that threat data in, analyzing it quickly using ML and AI, and then addressing those threats also using AI and more automation.”
— Chris Townsend, [37:14]

“The time to identify and resolve a threat can drop from hours to minutes using AI.”
— Chris Townsend, [41:20]

---

Memorable Moments & Quotes (with timestamps)

• “Mythos will be a step up from it. But we're already in the world where these LLMs are good enough to find the 10 year old vulnerability that you ignored in your firmware or router.”
  — Derek Johnson, [05:08]
• “The cryptographic ecosystem in and of itself… needs to consider transitioning to becoming more agile or responsive to the needs of the mission, not to the updates from an OEM.”
  — Philip George, [25:27]
• “An adversary is going to leverage whatever resources available. They don’t care about the rules, they're just going to go after whatever they want.”
  — Philip George & Moderator, [28:18]

---

Segment Highlights & Timestamps

[00:00–13:06]

Opening, Context, and AI Industry News

• Hype vs. real risk in AI for cyber; Anthropic and OpenAI product launches
• Federal “FOMO” around new cutting-edge models

[13:06–33:40]

Interview: Philip George (Merlin Group)

• How federal agencies are dealing with AI, data hygiene, identity, quantum transition
• Visibility as the “underrated tool”; specifics on building cyber basics

[33:47–43:40]

Interview: Chris Townsend (Elastic Public Sector)

• Public sector event takeaways, data operationalization, standardization, agentic AI in the SOC
• Specific examples from DHS/CDM and new managed SIEM offerings

---

Conclusion: Takeaways for Federal Defenders

• Visibility is the true “superpower”—asset inventory, data hygiene, and identity management are still the strongest lines of defense, even as the sophistication of attacks and tools grows.
• AI and open standards can and should enable, not replace, these fundamentals.
• Collaboration, both technological (open standards, interoperability) and organizational (cyber aligning with business), is key to staying ahead.
• Don’t neglect cryptographic modernization and post-quantum readiness, but tie those to broad, foundational cyber risk management.
• Modern SOCs need to evolve: being AI-enabled is about augmenting operators, not replacing them—embrace change.

---

End of summary.

Transcript

[Greg Otto]

In the age of AI, Federal agencies best defensive cyber weapon may be one they've had all along. We'll talk about it on this episode of Safe Mode. Welcome to Safe Mode. I'm Greg Otto, editor in chief at cyberscoop. Every week we break down the most pressing security issues in technology, providing you the knowledge and the tools to stay ahead of the latest threats, while also taking you behind the scenes of the biggest stories in cyber security.

[Interviewer (Cyberscoop Host)]

An attack is coming.

[Greg Otto]

It's about keeping us safe. He's just a disgruntled hacker.

[Interviewer (Cyberscoop Host)]

She's a super hacker.

[Philip George]

Stay alert.

[Chris Townsend]

Stay safe.

[Greg Otto]

Stay safe. This is Safe Mode. Welcome to this week's episode of Safe Mode. I am your host, Greg Otto. We have not one, but two interviews for you this week. One will be with Philip George, a technical strategist from Merlin Group. We talked with Philip out at the RSA conference about all things federal cybersecurity, AI, post quantum identity management. Really interesting conversation there. And then on the back end of the podcast, we'll be talking with Chris Townsend, the global vice president for Elastic Public Sector, talking about the data that goes into AI and how federal agencies can use that data to power their AI and really power their cybersecurity strategy. But speaking of AI, really heavy AI podcast, it's been a doozy of a couple weeks for AI and cybersecurity. Derek Johnson joining us. I don't know about you, Derek, but I just that the volume of responses that I got to Anthropic's project glasswing has really set off the industry, whether it's public sector, private sector, everybody wants

[Interviewer (Cyberscoop Host)]

to talk about it.

[Greg Otto]

So you've done a lot of reporting on it for us. Talk about what this week hath wrought for our audience.

[Derek Johnson]

Yeah, well, there's, I mean, it has been just like you said, a really crazy few weeks in terms of AI news. And it really kind of, I think all started off with the kickoff of PLOT Project glasswing, which is this anthropic led initiative to essentially take this new model that they have, Mythos, which they are claiming is too dangerous to release publicly or commercially. And they're essentially giving it to your Microsofts, your Cisco's sort of a small, a small group of about a dozen major tech companies just for the purpose of using it for cyber security. And so that's kind of kicked off a larger debate. Open AI, which has its own trusted access for cyber program that they've been running, very shortly announced a foray of their own, which is much, appears to be much more broader in terms of the number of people they're looking to get their tools in the hands of. But it's really, I think, accelerated. And for the mainstream, a conversation that we've been having over the, really the last three or so years about, you know, seeing the way that these large language models are getting better and better at cyber security specific tasks, but particularly, you know, things like vulnerability discovery, which can be in the hands of a defender, a very good thing, or in the hands of an attacker, a very bad thing. And so I think a lot of the debate we're having right now is how bad do we think it's going to get? Where is their hype? Where is their genuine concern? And I think that's kind of what we're working through right now.

[Greg Otto]

So speaking of hype versus genuine concern, there have been a lot of experts that have put the virtual pen to paper in terms of what they feel the future is going to look like. You did a story for us that examined that. What were some of the thoughts that came out in these papers that were released earlier in the.

[Derek Johnson]

Yeah, I mean, you saw. So it was really two, two really interesting papers. One was essentially a sort of a collection of a who's who of cybersecurity luminaries in the United States, foreign policymakers, high level folks in the private sector, CISO of Google, and sort of other people who have set policy or worked in the trenches at the highest levels. And their assessment was really that this is going to cause a major problem over the next few years simply because attackers are more position conditioned to use AI quicker, more recklessly, with less bureaucracy than defenders are. So even if you do have defenders adopting these AI tools, there's still going to be that kind of bureau, bureaucratic lag that legitimate organizations have to deal with. And then, you know, there's this, there's this long held sort of belief that it's going to completely sort of upend the vulnerability landscape. And I think that, you know, we just kind of had a conversation with some folks where, you know, they were saying it's not, we're not necessarily seeing the apocalypse. And you have to remember the models right now are pretty good at doing the things that Mythos and other folks are claiming they can do. So Mythos will be a step up from it. But we're already in the world where these LLMs are good enough that to find the 10 year old vulnerability that you ignored in your firmware or router. So.

[Philip George]

Right.

[Greg Otto]

And I think, excuse me, I think a lot of the context this week has also been centered around further context that yes, these bugs can be found, but they're bugs don't operate in a vacuum of course, even though that these vulnerabilities are there. I know for instance in the anthropic report there was like a 27 year old bug found in BSD which is a security focused operating system. And a lot of experts were like, well there are other conditions that need to be acted upon in order for this type of vulnerability to really be something along the lines of damaging to the level that we see of with your standard RCE vulnerability or there's other conditions that need to be met or that these bugs are so rare and so hard to act upon that yes, it's great that these bugs were found. But a big reason that these bugs weren't found by let's just say traditional static analysis or fuzzing or whatever is because if these bugs have been found, there is, there is always that human that goes yes, that is a bug, but the reality of this stuff actually being acted upon is just far fetched.

[Derek Johnson]

And maybe, and maybe that is changing. But I think the, the interesting thing about the second report which is from the AI Security Institute, it's the UK's top government's top AI research center, they actually got access to the model and were preview version of the model and were able to play around with it, subject it to capture the flag type tests and cyber range testing to get a sense for its capabilities. And there's conclusions were I think on the whole more measured than what we saw come out of the US side. I think they both pointed generally in the same direction of this is a real problem and Mythos is and models like it are a step up. But I think the gap between Volmageddon and you know, this is all hype. I think those two reports are really, really excellent to, to consult because they show you the, the bad future in the good future all in one.

[Greg Otto]

So speaking of hype from, I would say to, to borrow a term from other culture, I'd say sneaker culture or fashion culture. From hype to hype beast. Because I feel that the government, especially the federal government, we've seen them after this announcement go well wait, I want what, I want that too. Like can we play with Mythos as well? And we've seen reports that the government is very, very interested to apply this to their own systems or apply this to systems of critical infrastructure. And I, I don't mean that from the critical infrastructure in terms of ot. I just mean, what the government considers critical infrastructure, particularly the banking sector, that there's been reports that treasury is very, very interested to be like, hey, we want to play around with this because we want our constituency or what we oversee in terms of a regulatory aspect to make sure that everything is. Is buttoned up. But from the conversations that I had, I'm just the hypebeast part really stuck out. And I've never really seen that before, that hype beast in the cyber security landscape. Everybody's like, oh, that. That's the hot new thing. I want to play with the hot new thing. Why can't I have access to the hot new thing?

[Derek Johnson]

Yeah, yeah, well, and that's kind of, I think, gets to some of the differences between what we saw in Thropic rollout and what say, Open AI is doing. Because it kind of touches on both of those philosophies, right? One is like, you have this. So, you know, this Mythos model, which is supposedly so dangerous that you don't want to get it in the hands of too many people. Well, that would lend itself to sort of a more smaller, tighter, more centralized approach to using it. But OpenAI and you can say this is either a, you know, sort of a form of marketing on their end or whatnot. You could say the same thing for anthropic. But their position is this, essentially that, hey, we don't want to be the ones who sit here and decide this sector gets access to this tool, this sector doesn't, this company does, this company doesn't. We understand that some of that has to happen. But, but our goal is basically to get this tool in the hands of as many people, as many defenders as we can, so that they can use it to. To find bugs. So it's that. It's that kind of push pull of who do you give this to? How. How widely do you make that access? And. And what are the trade offs and repercussions for both sides?

[Greg Otto]

So let's dive in real quick on that Open AI part too, because we covered their announcement as well. Let's deep dive into what exactly did Open AI release because it was similar to what Mythos is going on. But there are some differences here.

[Derek Johnson]

There are some differences. So, you know, Mythos is. Is this brand new kind of frontier model that is separate. It's not a iteration or a variant of an existing model. It's a new, brand, brand new model at a higher level than. Than their previous ones. What OpenAI has released is a variant of an existing model that they have that has really been optimized and fine tuned specifically for cyber security tasks. And I know that they're not sort of making one to one comparisons between what they're, the tool that they're releasing for trusted access for cyber and what you, you know, what Mythos is doing. I think they see those as, as, as, as different, but in terms of where the program is right now, they're both using tools that in my estimation, just from the reporting that I've done, are probably good enough to do a lot with already. Right. You don't, I mean, you don't necessarily need Methos to make to, to find a lot of stuff that you could potentially fix or address in your network. You could really do that with the models that are out today. It's just that, you know, Mythos is going to be more powerful and faster and things like that.

[Philip George]

Great.

[Greg Otto]

Derek. Just a fascinating time. I say this every time, but because it just changes from week to week, I feel like always a fascinating time to be covering this stuff and how it applies to cybersecurity. And look forward to see how you continue to give our audience the best of what we know from the communities that continue to talk about this stuff.

[Derek Johnson]

Yep, absolutely not.

[Greg Otto]

To our interview segments for this week. First, we will be hearing from Philip George, executive technical strategist at Merlin Group and whether it's AI identity post quantum cryptography. Philip really is embedded inside federal agencies and is seeing what agencies are doing to protect their enterprises. And the big takeaway for him is really that cyber hygiene and particularly visibility is more important than ever. Even as we see a new influx of technology, the basics really still do matter and can still do wonders for agencies looking to protect themselves. And visibility is a key part of that. So we dive into that at all of the angles and all the technologies that I just mentioned. And then later on in the episode, we're going to be talking with Chris Townsend. I talked with Chris on the sidelines of the Elastic Public sector event we held last month. Really interesting conversation, particularly around AI and how data matters so much for AI. You really have to consider how well your enterprise is handling its data if you want to get the best out of AI as you integrate it into your enterprise. And Chris tells us all about what he's hearing from the public sector as they continue to put AI inside their enterprises. Check it out.

[Derek Johnson]

All right.

[Interviewer (Cyberscoop Host)]

Joining us on this week's Safe Mode is Philip George, the executive technical strategist at Merlin Group. Philip, thanks for joining us out here at rsa. I Know that even though we're all the way out here in California, a lot of your work is rooted in the federal government. So I definitely want to dive in to what you are hearing and the work that you are doing with federal agencies right now. So thanks for joining us.

[Philip George]

Yeah, thanks for having me. It's a pleasure to be out here. One, in our city and it's great. Two, just have the opportunity to engage. My role within Merlin as an executive technical strategist is comprised of multiple horizontal sip war interactions that these we are specifically focused, laser focused on acting as a catalyst or desired change across the public sector space.

[Interviewer (Cyberscoop Host)]

Okay.

[Philip George]

And also helping to provide innovation, innovative outcomes for the federal government, state and local organizations that kind of follow similar sort of regulatory structure and or higher security posture for their IT technology environments as that horizontal resource. I help the company as well as federal consumers understand these cases and where technology and those use cases align.

[Interviewer (Cyberscoop Host)]

Okay.

[Philip George]

Potentially even identify areas of greater efficiency.

[Greg Otto]

Okay.

[Philip George]

Especially in today's day and age, we're saying considerable focus on revitalizing tech spend and ensuring that the spend that's occurring today is providing the expected outcomes tomorrow and also identifying opportunities for greater collaboration and convergence of tooling and expanded outcomes with the reduce aces.

[Interviewer (Cyberscoop Host)]

So let's talk about those expectations, especially in a cybersecurity lens. Like, I'm wondering what you see in terms of the most common disconnect between cybersecurity investments inside the public sector and actual mission value.

[Philip George]

Yeah, that's a great question. When it comes to the actual investment, a lot of times the federal customer comes to us with a need and not necessarily an understanding of tooling. And so it's our job as members of the technology channel to reconcile that need with capabilities across the technology ecosystem, finding ways to introduce emerging and or potentially disruptive tech in a safe but secure manner for consumption, or what we like to call suitable for federal service.

[Interviewer (Cyberscoop Host)]

Right.

[Philip George]

Okay. And you know, as we all know, the federal government has a higher level of standardization around how it consumes technology. It also has concerns around the supply chain angles around where that technology is produced in a pedigree thereof. And so Merlin tries to buy down that risk while maximizing clear and transparent deliver deliverables of those tools into federal spaces.

[Interviewer (Cyberscoop Host)]

So even on top of cybersecurity with being out at rsa, you can't go five feet without hearing about AI. And I know from talking to a bunch of the experts that we talk to in the federal space that AI and ML adoption is something that agencies are experimenting with the Same way that the private sector is experimenting with it. And I know that one of the challenges, particularly for the federal sector, is how the data gets used because there's so much data inside these federal agencies that drive the mission really. So on top of the cybersecurity issues and the mission that comes with protecting that data, can you walk us through what data hygiene conversations you're having, especially through the lens of AI and ML adoption? Because I do feel like this is such an interesting conversation that is so federal centric right now with talking about data governance and data sovereignty and how that all factors into this AI boom that we're seeing.

[Philip George]

Yeah, so another great question. It's a perfect storm of need, exposure, risk and potential mitigation of said risk. You know, federal organizations have for a very long time struggled with conducting proper identity management, data hygiene curation and just understanding of their, their data environment, their data lakes for that matter. And so none of that has changed. But now we have a scenario where we're plumbing access to these data lakes, data repositories, and providing said access to a highly privileged, highly capable large language model and or AI tool for that matter. And the tooling and the know how around how to do that safely and securely is still lagging behind. And so where we try to partner with our federal partners is to identify ways to leverage some of the more novel AI products and helping them map out their data, the structure of it, the quality of it, as well as who's accessing it. And a lot of that starts with having sound identity management around for human, non human purposes, being able to analyze behavior of those identities and the usage of those identities and identify benign versus malign activities. And so as we're able to build kind of like the guardrails around AI usage, we're also able to enable and continue to support rapid adoption of those AI tools. And so let's start at the foundation be sounding your application of identity management products as well as data integrity tools, because that's the other part of it too. You want to ensure that the quality of your data is sound enough to start to rely on these automated decisions while still keeping the human in the loop. So integrity of ops, the integrity of accounts and visualization of who's doing what and where is, are what we like to call some of the key first first steps in.

[Interviewer (Cyberscoop Host)]

So let's dive into a little bit of those first steps. What are some of the specific vulnerabilities that emerge when agencies are rushing through an AI adoption without understanding the attack surface that can expand.

[Philip George]

Yeah, absolutely. So the, the biggest risk is over, over provisioning access into data environments, right?

[Interviewer (Cyberscoop Host)]

The identity part.

[Philip George]

Yeah, the identity part. Once again, if you have poor tooling, logging around who's accessing what and in what method and or manner for that matter, you may or may not, you know, be losing data through exfiltration because of the fact that you open up the aperture to automated accounts, non human accounts, service accounts, and you can't tell very quickly or effectively what's legit, what's not legit. A lot of the times our adversaries like to live off the land and use legitimate accounts in illegitimate or malicious ways to surreptitiously access data, change permissions and, or potentially even exfil information right under our nose. And so it's easier to leverage existing accounts, existing privileges, than to establish their own tools and kits for that matter, to exploit a given target. And so that's one particular threat vector, you know, protecting the prompt, being able to identify data poisoning, you know, if we are making some critical decisions, whatever the market or industry may be, critical infrastructure or financial data for that matter, you have to make sure that the data that's being used for these automated outcomes is of high quality and high integrity changes can be reconciled. And if you can't reconcile those changes, then maybe you should question some of the automated outcomes prior to taking action.

[Interviewer (Cyberscoop Host)]

So with all of that, I'm sure there are some federal CISOs out there or some security leaders that might be hesitant to rush into this. How should federal CISOs navigate this when faced with executive pressure to deploy AI capabilities quickly?

[Philip George]

Yeah, so that's, it's a tough question, but one that, you know, every CISO is grappling with. I, I always recommend working with trusted partners, organizations that, you know, have less of an interest to, you know, utilize labor or burn labor hours. More of an interest to provide you with the right solution that meets your specific needs, buys down risk, mitigates some of the, some of these very issues that we're discussing today, while at the same time enables the business unit. Cyber in and of itself is an enabling function. It does not exist to serve itself. It exists to equip and enable the business. And so aligning some of these cybersecurity capabilities with clear business drivers helps to bring both entities together as a partnership, which will ensure success and continued interaction along their AI journey and operational deployment. So that's one of the recommendations. We say both teams need to work together in unison and lay out some achievable low hanging fruit outcomes around integrity of data. What is the high watermark or the high impact element of your business operations and how do we protect and or achieve that? And so yeah, building consensus, targeting low hanging fruit from a cyber outcome perspective and winning over the business to obtain their confidence that the cyber team can deliver. Not through saying no, but this is the way, this is how we do it versus we can't do it.

[Interviewer (Cyberscoop Host)]

So I want to switch gears a little bit. Listeners of our podcast a few weeks ago heard from Garen Lacy, a State Department official, really talking about the need to think about post quantum cryptography and quantum security and really calling out to industry to say, hey, we need to get on board with this. We need to start now before it's another time honored tradition of the government playing catch up.

[Philip George]

Yeah.

[Interviewer (Cyberscoop Host)]

So given that quantum computing threats still might be years away, how should federal agencies be prioritizing against this and planning for this when they still have to deal with the more immediate threats? What's the rights balance?

[Philip George]

Yeah, so you know, it goes once again back to recapitalizing your tech investment. There's not a lot of new money to solve some of these new challenges, but you know, we, this is an opportunity to be somewhat reflective and look at where that spend is occurring. And is that spend paying dividends towards new and emerging security outcomes? In the area of post quantum cryptography, there's quite a bit of concern and need to modernize the cryptographic substrate and ecosystem. It's a very legacy ecosystem, but it's also very pivotal to almost every facet of technology delivery today from a confidentiality, integrity and availability perspective. And so CISOs and system owners for that matter, definitely need to overlay cryptographic modernization, potentially starting with a cryptographic inventory and discovery of their assets. Cyber 101 is to have a comprehensive and an accurate inventory of the assets that you have to protect. And so if you're considering or you are subject to some of the CNSA 2.0 or M2302 requirements, then it's incumbent upon the system owners to begin that journey by conducting that automated inventory, discovering where cryptography exists, not just what's in use, but what's available, and correlating that with what's actually being seen over the wire. So you have this comprehensive input and visualization of how crypto is used in your environment. You'll discover classical vulnerabilities as well as the potential post quantum vulnerabilities from a harvest now decrypt layer perspective, all of which are going to kind of lay out Your migration and remediation roadmap. Furthermore, there's a lot of change in the PKI ecosystem. Certificate Life cycles are being shortened. Why? Because their persistence only encourages additional living off the land attacks. And so if our adversaries are able to censor and discover some of these assets, they once again can falsely or maliciously authenticate as a legitimate process and hide in our noise floor. And so in order to reduce that noise floor and that potential exportation vector, we have to reduce the longevity of these certificates and it's only going to continue to get shorter and shorter. And so the cryptographic ecosystem in and of itself has to be characterized and it needs to consider transitioning to becoming more agile or responsive to the needs of the mission, not to the updates from an oem. And so this is how we operationalize that cryptographic ecosystem and add it to the cyber risk management cycle and conduct what we like to call cryptographic posture management effectively throughout the cyber apparatus.

[Interviewer (Cyberscoop Host)]

So I want to go back to something that you said there really about visibility. It's something here at rsa, whether I'm talking about the public sector or the private sector, when we're talking about new technologies, whether it is quantum security or AI, agentic AI, whatever, everything goes back to, from a defensive perspective to I'm surprised that the conversation has always gone back to visibility. That like that alone can quell a lot of the worry about what is capable from a threat perspective.

[Greg Otto]

Do you feel the same way in that?

[Interviewer (Cyberscoop Host)]

I just find it fascinating that everybody goes, actually, do you know what the best way is to defend against itself? Figure out what's on your network from square one and then the rest of the problem will start to take care of itself.

[Philip George]

Yeah, well, you can't protect what you don't know. And you know, from a human perspective, we're very visual creatures and we also now live work under the auspices and guidelines of the Zero Trust concept. We're no longer supposed to implicitly trust the tools, the environments, the products that we're consuming. Whether it's deployed by ourselves or provided by a managed service provider, we have to verify, and we must verify everything in order to do that. You have to be able to inventory and visualize it in the first place. And so yes, having an authentic, comprehensive inventory of your assets, be they devices, cryptographic assets, or AI tools for that matter, is important for you to establish your defense in depth posture and strategy. There are numerous instances of rogue AI agents, as well as custom cryptographic algorithms being deployed for test purposes. That doesn't Mean that Navisary won't mess with the test environment. An adversary is going to leverage whatever resources available.

[Event Host/Moderator]

Right.

[Interviewer (Cyberscoop Host)]

They don't care about the rules, they're just going to go after whatever they want.

[Philip George]

Yeah. So we have to increase the cost of their operations by raising the bar around of scrutiny and visibility of all of our privileged assets and non privileged assets so that we're better able to mitigate that attack surface, reduce that attack surface and get left the boot. That's, that's the goal of0Trust PQC migration and agentic AI adoption is to move less towards, I mean move more towards going left of attck and preventing it by looking at the forensic effects of said attack. And so we're on the wrong side of that equation. Nine times out of 10, how do we change that? Well, we shift the environment and that's really what AI PQC is doing today. It's shifting that cryptographic substrate. It's shifting how effectively and efficiently we can contextualize large volumes of data, make sense of it from an information perspective, and then take definitive action from a cyber defense and potentially offensive perspective as well too.

[Interviewer (Cyberscoop Host)]

So you brought up Zero Trust there and I've been thinking about this a lot. We've been reporting about it. I feel like maybe you don't share this opinion. I'll just throw this out here.

[Greg Otto]

Zero trust in agentic AI.

[Interviewer (Cyberscoop Host)]

I feel like are at, in like combat with one another. I feel like if you have an agentic AI that you know, you want to deploy internally, to go do this, that and the other thing.

[Greg Otto]

Yeah, there's like they can get unwieldy.

[Interviewer (Cyberscoop Host)]

Like we've seen what, what has happened with just open claw in the, the private sector. I'm sure there are some federal agencies that are also experimenting with openclaw too. I'm sure that probably gives sizzos a headache to, to hear that, especially when you think about it from a zero trust perspective. So I'm wondering that Zero trust in AI are sort of like at loggerheads with one another or is there, can there be harmony there?

[Philip George]

I guess, yeah, I think there can be harmony. But I understand why there's discourse and mainly it's because, you know, AI just exacerbates all the challenges that we've had with just human processes.

[Interviewer (Cyberscoop Host)]

Right, right.

[Philip George]

And so it's accessing everything, it's potentially changing things. It's making decisions that we may or may not be able to contextualize or characterize within a reasonable amount of time. And so we're getting overwhelmed. We're getting overwhelmed with logging data flows, modifications and the whys. And so if a given cyber organization is struggling with human processes, they may very well get overwhelmed with some of these agentic AI processes and workflows. And oddly enough, I think the answer is a little more AI. Okay, right. And you know, there are a lot of tools that are out there that help you do a better job of making sense of your logging activities, making sense of, once again, account identity, account usage, human, non human account usage. And this is where the cyber team is going to have to get off of their, their old favorites and use some new and emerging tools that are very narrow in scope but clear in outcomes to kind of ingest and deal with the, the influx of activity and output from the business processes. And so sadly, well, not sadly, but oddly, the answer is a little more purpose built AI to kind of enshrine and protect the larger organizational AI use cases and outcomes. So partnering Key Cyber has some clear support outcomes, but the business has the larger priority outcomes. Both need to move forward in an effective manner.

[Interviewer (Cyberscoop Host)]

So to sum this all up, if you could mandate one change across all federal agencies tomorrow, whether it's policy shift, technology investment, a change in approach, what would have the biggest impact on our collective security posture?

[Philip George]

Yeah, I would probably try and standardize how we, you know, and protect AI usage. Okay, Identity usage, for that matter. There are a lot of solid tools that are out there today that go beyond just the, the PAM password account management where we're, you know, vaulting authenticators for that matter, but also building behavioral analytics around these accounts, these human and non human accounts and how they're being used, who's using them, and is there a potential indicator of compromise? So I think I would start there. I'd also make it a priority for federal organizations to put more time and effort into understanding the quality of their data and enabling their data officers for that matter, to be a little bit more authoritative in ensuring that the data that we are aggregating and utilizing has integrity checks and all the requisite auditing around it to ensure that no poisoning is happening. And all the changes can be reconciled from an organizational perspective as well too. So identity, data integrity, those are probably the two areas that I would prioritize from a federal perspective to see more rigor and adoption of effective AI tools around.

[Interviewer (Cyberscoop Host)]

Great, Philip, really appreciate your time. Thanks for hopping aboard the program.

[Philip George]

Thanks for having me.

[Event Host/Moderator]

All right, and joining us for a special segment on Safe mode work. We're coming to you live from the Elastic Public Sector Summit.

[Greg Otto]

And I'm here with the man of

[Event Host/Moderator]

the hour, Chris Townsend, the global VP of Public Sector for Elastic. Chris, great event today. What were some of the takeaways that you saw?

[Chris Townsend]

Thank you so much, Greg, and pleasure being here with you today. So there was a number of really interesting takeaways from the public sector event today. So first off, it was very well attended, which I was pleased about because obviously there's a lot happening in the world right now. A lot happening with dhs, a lot happening with dod. So as really pleased to see the turnout that we had across all of public sector, both federal and state and local. Couple of takeaways. Everyone's trying to figure out how to operationalize data, right? Agencies that are going to succeed going into the decades ahead are going to do it by putting their data to work for them. And a lot of the conversation today was around that. Really three areas that we focused on. One was agentic AI. Talk more about that. The second was cybersecurity and the evolution that's happened in cybersecurity. And the last was really reducing the cost of of data to infrastructure. The final takeaway is these challenges are really complicated. If you look at agentic AI, you look at the evolution in cybersecurity and we're seeing more and more collaboration between government and industry. That's really refreshing because these problems are hard to solve in silos and we're breaking through those silos and really collaborating more. And I think that's a lot of what today was about.

[Event Host/Moderator]

So on the cybersecurity side, especially when it comes to all the things that are going on, we just saw the President release a cybersecurity strategy. How do you see agencies leaning into cybersecurity, especially on the modernization or the resilience front? Because the threats are speeding up. So being resilient is definitely a top priority for these agencies.

[Chris Townsend]

Yeah, the threats are speeding up. And look, it's always been cheating Antelope scenario. We get a little bit ahead and then the adversary scene catch up, but that's changing. Our adversaries have access to these AI tools as well. And the adversaries are evolving much more quickly than they have in the past. And I really worry about some of our government agencies in particular because of the procurement cycles and the evaluation cycles. They have legacy technology that doesn't position them well for that fight. So we need to help our government agencies modernize quickly. And a lot of that modernization is about embracing open standards and getting tools to work together. There's lots of security sensors, if you will, out there. There's endpoints and networks and a lot of different components. Those sensors need to communicate better and share threat data. And then we need to layer I layer AI on top of that threat data to do threat hunting or to do SOC automation. And we need to really help our security analysts pivot from trying to react to being more proactive and shutting down threats quickly as they occur or before they occur. And AI is helping us do that. And that's not something that's kind of future pie in the sky. That's in practice today. We're seeing that.

[Event Host/Moderator]

So before we jump into the AI part, I really want to hit upon the standardization part because I heard a lot about that on stage. And while the standardization part will never be as sexy as the AI part of it, that is a really important part, especially when you're talking about cross agency collaboration or what needs to happen when civilian agencies are talking to one another and sharing threat data. So can you kind of take us behind the scenes on what Elastic talks about when it is talking about standardization, how that applies to data, how that can ultimately help agencies achieve what they need to achieve?

[Chris Townsend]

Yeah, great question. If you think I've been in cybersecurity for a long time and if you see how cybersecurity was built out, it tends to be very reactive. Right? Is a threat to a threat pops up or a new threat vector pops up, you buy a new tool, you deploy that tool. Often those tools are running proprietary protocols, they're siloed, they don't communicate with the other tools. I think the industry has realized collectively that security is a data problem and it's really about bringing that threat data in, analyzing it quickly using tools like ML and AI, and then addressing those threats also using AI and more automation. Elastic started out as an open source company, so we're very much an open source and open standards based company. There's a project out there called the OpenTelemetry Project. They approached us about a year and a half ago and asked us to open source all of our data protocols for observability and security, with the goal of facilitating data sharing not only between data tools, but also security tools themselves and how they generate logs so that we're able to have a common schema that allows us to do the data analysis on those more effectively and allow that data to be shared. So we're seeing a focus now, especially across the federal government, to embrace more of those open standards. And a good example of that is we talked about this a bit on stage today is Department of Homeland Security for the last seven or eight years has had a program called the Continuous Diagnostics and Mitigation Program. And essentially the CDM dashboard set a baseline for the federal civilian agencies to be able to raise the minimum standard up for cybersecurity and then monitor the cybersecurity events that were happening in those agencies. They use Elastic for that dashboard. So there's about, I think, 97 agencies that fall within that continuous diagnostics and mitigation program. They have small elastic clusters that are embedded in all those agencies that pull data and they're allowed to. What happens is the CDM dashboard that allows them to query that data, where that data lives. So it's not replicating all the data, it's not duplicating the data. It's. It's indexing the data where it is. And they're able to look for indicators of compromise or threats within those agencies. But do so while the agency is using their own tools and data sets. What the SIM as a service program is doing is expanding that capability. And essentially CISA is saying, hey, you know, we've been using Elastic to do this monitoring. We're now going to offer Elastic as a managed SIEM platform. So if you want to take all of your security data, your threat logs, and we will deliver Elastic as a platform, as a service partnership with ecs and the agency can maintain their own SOC operations, folks, but they would essentially use Elastic as a platform, as a service. What that does is facilitate data sharing across all of our civilian agencies. And it gives CISA visibility into all that data so that they can also augment those capabilities of the agency to do threat hunting across a much broader spectrum.

[Event Host/Moderator]

And then you get onto the AI part of it. You have all of those other tools that you just talked about, but then, you know, we're talking about putting AI on top of everything else. And look, with AI, we've seen it really speed up what is possible in both attackers and defenders. Like, it's totally changing the game when it comes to cybersecurity. So what are you hearing from customers that you're talking to on the ways that they're utilizing AI to get the

[Greg Otto]

best out of it?

[Philip George]

Yeah.

[Chris Townsend]

So there's really two key components there. We've had machine learning for a long time and some rudimentary AI to do threat detection and threat hunting on unknown threats. And that's continue to advance with, with all AI and we've got some great results and things that we're doing there. What's really been a step change over the last 18 months, two years, has been the AI enablement of the SOC operator. So if you're an analyst and you come into the SOC in the morning and you've got 200 alerts that you have to parse through and figure out which of those alerts are real, which is noise, which are correlated, you know, which need to be prioritized. That's what your analysts do every day. Now we've built in Agentic AI into the SoC, so we've enabled the SoC with an AI tool. And this just isn't an AI assistant, it's an LLM that's purpose built and trained with the mitre, ATT and CK framework, all of the things in the head of your level one, level two SOC analysts. So now they can take those alerts, put them into the LLM. The LLM will then prioritize them. It has unique knowledge of your environment, mitre, ATT and CK framework, your runbooks, playbooks. It will prioritize those threats. And not only will it prioritize which threats are real and targeted at your organization, it'll also then give you the step by step process to remediate those threats. So the time to identify and resolve a threat can drop from hours to minutes using AI. And you think about how difficult it is to retain and train and maintain your SOC analysts. It's a very tight supply of SOC analysts. The fact that we can make the SOC analyst their lives better, make them more efficient and effective and also take your more junior SOC analysts and make them operate at a higher level because you're bringing in AI, right? And that's really the promise of agentic AI is a partnership with a human, not a replacing.

[Event Host/Moderator]

So you show what the art of the possible is there. And if you attended this event and you go back into an agency, I'm a security analyst inside the government. What conversations do I need to be having to make sure that when we're a year out from now that I can realize what the art of the possible really is and really get to my mission and defend my mission if I'm inside an agency? And that's what I'm worried about.

[Chris Townsend]

Yeah, and that's, that's a really great concern to have because you know, there's these security platforms as we talked about earlier, some of these security platforms that really aren't up to where we need to go to prevent against these AI powered threats. They're deeply embedded in these very large agencies. Folks are trained on them. They've got custom dashboards, they're very large enterprise software deployments. So to migrate off of them is hard and change is hard. So I would say the number one thing if you're a SOC operator is you have to have an open mind and be willing to, hey, if you know the query language and how to run this platform and you've been doing it for years, be open minded that hey, there may be better things out there and you have to explore what else is out there and not just get locked into things that you're familiar and comfortable with. Because look, everybody dislikes change. And what we find a lot is just because folks have a familiarity with an older tool, they want to stick with that older tool and it's, and it's slowing down modernization of some of

[Event Host/Moderator]

these Great, Chris, great event, great talking to you. Really appreciate you joining the program.

[Chris Townsend]

Oh, thank you for having me. Appreciate it. Thanks Greg.

[Greg Otto]

Thanks for listening to Safe Mode, a weekly podcast on cyber security and digital privacy, brought to you by cyberscoop. If you enjoyed this episode, please leave a rating and a review and share it with your friends, your co workers, your CISOs, your sysadmins, your mom, your dad, anybody that wants to know more about cyber security. To find out more information or to contact me, please look for all of our social media handles or visit cyberscoop.com thanks for listening. Check us out next week.