Chloe Maraina brings a rare blend of data science rigor and business intelligence vision to the rapidly shifting landscape of cybersecurity. As an expert who spent years deciphering complex datasets to find the “signal in the noise,” she has a front-row seat to the emergence of agentic AI—autonomous systems designed not just to flag problems, but to think, navigate, and remediate in real time. In this conversation, we explore the “Mythos moment,” a turning point where AI-driven vulnerability hunting is moving at a speed that threatens to overwhelm traditional human-led IT teams. We delve into the growing tension between the undeniable efficiency of these tools and the deep-seated skepticism of enterprise leaders who are wary of letting autonomous agents run wild in their sensitive infrastructure.
The themes we cover include the logistical nightmare of a “patch avalanche” caused by hyper-efficient AI scanners, the practical reality of transitioning from “crawl” to “run” phases in AI adoption, and the new “identity apparatus” required to manage a workforce of digital agents. Maraina provides context on how platforms like Cisco IQ and Cloud Control are attempting to bridge the trust gap by offering “scary good” diagnostic capabilities while keeping humans firmly in the loop.
AI models like Claude Mythos can identify thousands of high-severity vulnerabilities in mere weeks, a speed that feels almost predatory. How is this “machine speed” mapping fundamentally changing the risk profile for the average enterprise?
The sheer velocity we are witnessing right now is enough to give any CISO a sleepless night. When you look at the Project Glasswing preview, partners unearhed more than 10,000 high- or critical-severity security flaws in commonly used applications within just the first month. We aren’t just talking about a slight uptick in efficiency; we’re talking about more than 1,000 open source projects being picked apart for weaknesses in thirty days. This creates a terrifying window of exposure because, as Liz Centoni pointed out, attackers aren’t waiting for the general availability of these tools to start their own scans. They are already out there, mapping networks and finding end-of-life devices in minutes, and that forces enterprises into a defensive posture that is reactive by default. The risk profile has shifted from “can we find the hole?” to “can we plug it before the automated hammer hits?” It feels like a race where the opponent has a jet engine and the defenders are still trying to lace up their running shoes.
While the diagnostic power of these AI agents is often described as “scary good,” there seems to be a significant psychological barrier to letting them act autonomously. Why is the “trust gap” proving to be such a persistent hurdle for even the most tech-forward organizations?
Trust in the enterprise isn’t built on a successful demo; it’s built on years of predictable outcomes, and right now, the “move fast and break things” mentality is a non-starter for infrastructure. When a senior network engineer like Don Cheney from Washington Trust Bank calls something “scary good,” it’s a compliment to the tech, but it’s also a warning about the lack of control. These engineers are responsible for the lifeblood of their organizations, and they know that if an autonomous agent makes a wrong turn, everybody suffers. There is a deep-seated fear of sensitive data exposure and the “black box” nature of AI reasoning, which is why we see a conservative “crawl, walk, run” methodology being adopted across the board. Even when an AI provides a perfect step-by-step troubleshooting guide for a Meraki dashboard, there is a visceral hesitation to let that AI pull the trigger on a fix without a human eye on the screen. It’s a fundamental conflict between the need for AI’s speed and the human requirement for accountability.
The phrase “patch avalanche” has been used to describe the aftermath of these hyper-efficient AI scans. When an organization is buried under thousands of new critical vulnerabilities, how do they move from discovery to remediation without paralyzing their operations?
The reality is that finding the flaw is now the easy part, while fixing it remains the grueling bottleneck. We are facing a massive patch avalanche where the sheer volume of vulnerabilities exposed by models like Mythos can overwhelm even the most robust IT departments. Organizations are realizing that they won’t get a second chance if they don’t move decisively, yet they can’t reboot their entire data center every time a new flaw is discovered. This is why stopgap measures like Live Protect are so vital; being able to apply compensating controls on something like a Nexus 9000 switch without a reboot is like applying a high-tech tourniquet while the patient is still moving. It allows a company to prioritize their response and manage the onslaught without bringing their business to a grinding halt. You have to be able to triage with surgical precision because you simply cannot fight every fire at once when the AI is lighting them at machine speed.
We’re seeing a shift toward “agentic identity,” where every machine and service is assigned its own identity apparatus. What does this mean for the future of governance and how we define “who” or “what” is acting on our networks?
This is perhaps the most profound shift in the architecture of the modern network, moving us toward a world where humans are no longer the primary actors. As Jeetu Patel suggested, we are entering an era where every agent, service, and machine needs a rigorous identity apparatus, much like a human employee would. This isn’t just about security; it’s about visibility—knowing which AI agent is associated with which human manager and what specific tasks they have been delegated to perform. With tools like the Agent Gateway, we are starting to see the infrastructure for this, allowing for DNS-based discovery and role-based access permissions that are scoped separately from human credentials. It creates a digital paper trail for autonomous actions, which is the only way we can hope to govern a network that is increasingly populated by non-human entities. Without this identity layer, the “agentic era” would be nothing but chaos, a swarm of invisible actors making changes without any accountability.
With platforms like Cisco IQ and Cloud Control now in the hands of thousands of customers, what specific “GPS-like” capabilities are proving to be the most disruptive to the traditional way of managing networks?
The shift from a “static map” to a “live GPS” is all about moving from visibility to actionable intelligence. For the 2,000-plus customers who jumped on Cisco IQ within its first five weeks, the most immediate impact has been the automated asset inventory that sniffs out end-of-life devices before an attacker can exploit them. It’s about the system knowing what network it’s in—like the experience at Washington Trust Bank where the AI didn’t just point to documentation but provided site-specific, step-by-step troubleshooting within the Meraki dashboard. This level of contextual awareness means engineers aren’t spending hours mapping out the problem; the “Deep Reasoning” mode validates its logic against real-time telemetry data before the human even joins the session. It turns a technical support request into a guided mission, significantly speeding up the time to resolution and allowing the human operators to focus on high-level strategy rather than the minutiae of diagnostic digging.
What is your forecast for the evolution of AI-driven cybersecurity over the next two years?
In the next twenty-four months, we are going to see the “trust gap” begin to close, but not because of a sudden change in human nature; it will close because the sheer volume of threats will make human-only defense impossible. We will move away from a world where AI is a “scary” assistant to one where “DefenseClaw” and similar sandboxed environments are standard operational procedure for every transaction. We will see the maturation of agentic identity management, where the acquisition of companies like Astrix Security will result in a world where your AI agents have more audited credentials than your junior analysts. The state of Indiana’s 148% ROI on observability tools is just the tip of the iceberg; soon, the metric of success won’t be ROI, but “resilience time”—how many seconds it takes for an agentic defense to neutralize a Mythos-level threat. The organizations that thrive will be those that embrace the “Agent Gateway” model, successfully delegating the “machine speed” tasks to AI while keeping their human experts focused on the ethical and strategic guardrails.
