Cisco incorporating AI into cyber defense
An interview on AI, the risks, and the future of cybersecurity
Project Glasswing shows how cybersecurity is changing in the age of Frontier AI. In this interview, Cisco talks about the insights gained from working with leading AI models, why traditional vulnerability assessments are reaching their limits, and why identities, automation, and continuous defense will play a central role in the future. Cisco also explains which specific measures businesses should take now and why partners like NTS are crucial for successfully putting new security strategies into practice.
The biggest surprise from Project Glasswing
What surprised you the most when using Frontier models compared to your own products? More particularly: was it the number of findings, or the kind of insights? For example: Did the model link vulnerabilities to attack chains that, when considered individually, would each have had a lower CVE score and therefore would never have been prioritized?
The surprise is in the models’ “cognitive” performance. While conventional scanners find isolated vulnerabilities, Frontier models understand the context. They can recognize that an insecure configuration, combined with a slightly outdated library and a lack of segmentation, opens a path to a critical server. The key insight is this: The sum of the parts is more dangerous than the individual part.
Frontier AI collaboration in practice
Now that you have access to Frontier models: What exactly are you doing differently in your product development? A specific example from the past few weeks would be ideal.
We have taken the “shift-left” principle to the extreme. Through project CodeGuard, AI agents that assist with the software development are trained to apply “secure-by-default” rules. This means that we are trying to prevent security vulnerabilities from arising already in the IDE (Integrated Development Environment), rather than just scanning for them after compilation.
Volume versus value
Both vendors explicitly emphasize that raw vulnerability counts are not very meaningful, and Cisco speaks of “actionable precision at scale, not the count of vulnerabilities alone,” while Palo Alto Networks measures success based on exploitability and business impact rather than the number of CVEs. How exactly can a customer tell whether a scan result is truly actionable?
Cisco has moved away from simply counting CVEs. A CVE score says little about the real world. We now use Cisco IQ to determine whether a vulnerable feature on a device is even accessible. If a device is not connected to the internet or the affected feature is disabled, the priority is lowered. This conserves resources and protects teams from “alert fatigue.”
“Patching needs to become as boring as possible.”
The 3 specific things
If a customer’s IT manager could only do three things over the next 90 days to be better prepared for AI-accelerated attacks, which three would you recommend?
The 3 specific things (90-day plan):
- Inventory: without a “single source of truth” across all assets (including shadow IT), defense is impossible.
- Segmentation: if you cannot patch everything immediately, you must limit the area of attack through segmentation (east-west traffic control) to the point where an attacker cannot progress further if they gain a foothold.
- Routine: patching must become as “boring” as possible. Cisco’s 7-day advance notice allows customers to plan their maintenance windows precisely.
Identity as the new frontline
Both vendors are currently investing heavily in identity security for people, machines, and AI agents (Cisco/WideField, Palo Alto Networks/CyberArk). Why is identity, of all things, the key lever right now, rather than traditional patching?
Why identity? Because attackers today operate using a “living-off-the-land” approach. They do not use malware, but rather legitimate login credentials. Once an attacker is inside, they move around with valid privileges. Securing identities, especially non-human identities such as service accounts and AI agents, is the only way to stop this movement.
AI as a double-edged sword
Attackers are using the same technology. In your view, what is the biggest misconception customers currently have regarding “AI threats”?
The threats are well-known (phishing, SQL injection, credential stuffing), but due to AI the speed and quality of execution have increased by a factor of 100. We do not need a truly new defense, but rather a faster and more automated version of our existing defense. “You cannot outpace AI-era threats using legacy processes or legacy technology. What is needed is a continuous operating rhythm […] patching becomes an ongoing rhythm instead of a fire drill.”
A new dimension, a new time frame
The challenges customers are facing have reached a new dimension. Understanding and defending their own area of attack, applying patches promptly, and preparing for emergencies (incident response/SOC). All of this must be done within an unprecedentedly short time frame. What role do system integrators such as NTS play in ensuring that these developments actually reach the mutual customers in a timely and effective manner?
The complexity of the Cisco architecture (Cloud control, Splunk, ISE, firewalls…) requires experts who can piece these puzzle pieces together to a whole thing. NTS would act here as an “operational partner,” ensuring that the data provided by Cisco (from Cisco IQ) is actually translated into automated workflows (SOAR) so that the customer not only sees the data but also takes action. Partners like NTS are encouraged to establish recurring services related to assessment, hardening, and validation (using the Foundry Security Spec as a foundation). As Cisco so aptly put it: “A partner bridges the gap between identifying a risk and neutralizing it.” NTS assumes the role of the “unified defender,” which not only supplies hardware but also consistently applies security policies across the entire multivendor environment and manages the customer’s “assess, act, optimize” workflows.
Outlook
What do you expect to see by the end of the year? Will the situation ease or continue to escalate, and what does that mean for the collaboration with NTS?
No, it is actually going to get more intense. We are moving into an era of “continuous defense.” The collaboration with partners like NTS will evolve so that they no longer just supply hardware but act as “Managed Security Service” partner, who take over the operation of the security infrastructure, allowing the customer to focus on their core business.
“We are moving into an era of ‘continuous defense’.”
“8 years of security research in 8 weeks”
What does this statement mean specifically, and what activities did Cisco actually carry out as part of Project Glasswing (scope, approach, models used), and what are the concrete outcomes out of it?
This was a massive undertaking. We did not just scan codes, but we trained the models to understand our specific Cisco frameworks. By running the models in parallel, we were able to drastically reduce the time it would have taken a human team to perform the analysis manually. The result is a “Foundry Security Spec,” which we have published as a blueprint for the entire industry.
From findings to prioritization
Which conclusions does Cisco draw from these findings, and which topics are being prioritized as a result? Live Protect is one example. More generally: How are the insights from Project Glasswing impacting your short-term product roadmap?
The roadmap is now “risk based.” We are investing heavily in the automatization of the validation. When we find a vulnerability, the system must automatically check whether protective measures (such as firewalls or segmentation) are already in place. If not, it is prioritized automatically.
Live Protect as a bridge instead of a patch
You explicitly state that “Live Protect is not a patch,” but rather a temporary, targeted shield directly within the NX-OS kernel. How does this change the customers’ expectations regarding patch cycles? Does it replace a rapid patch cadence, or does it actually reinforce the need to patch in the long term anyway?
Live Protect is a “compensating control measure.” In the world of cybersecurity, this is a crucial difference from a traditional software patch. While a patch permanently removes or corrects the faulty code in the program, Live Protect places a targeted shield directly within the kernel (e.g., in NX-OS on Nexus 9000 switches).
Why is it a “bridge” and not a replacement?
- Bridging the “exploit gap”: The window of time between the discovery of a vulnerability and its exploitation by AI-powered attackers has shrunk to just a few hours. However, a traditional patch cycle (testing, validation, deployment) often takes days or weeks. Live Protect closes this gap immediately, without requiring a reboot or a service interruption.
- Not a permanent solution (time-bound mitigation): Cisco explicitly states: “Live Protect is not a patch.” It is a temporary measure. As soon as the official, permanent patch for the vulnerability has been applied and verified, the shield is automatically removed. Leaving the shield in place without a patch would create “technical debt” and make the system vulnerable to configuration drift.
- Avoiding “fire drills”: Live Protect relieves IT teams of the pressure to immediately switch to emergency mode (fire drill) every time a new security alert is issued. Customers can neutralize the vulnerability immediately and integrate the actual patching process into their regular, scheduled maintenance cycle (e.g., the first or third Wednesday of the month).
The strategic shift in patch culture
Live Protect fundamentally changes expectations regarding patch cadence:
- From reactive to controlled: customers no longer have to drop everything every time a notification is issued. Security is guaranteed by the shield while the patching process runs smoothly and securely.
- Automated life cycle: the documentation describes a process for managing the shield: “monitor, enforce, disable, and retire.” This means that Cisco not only provides the protection but also helps to remove it once it is no longer needed.
- Architecture expansion: while Live Protect is currently available on the Nexus 9000 (NX-OS), the plan for 2027 is to expand this technology to catalyst switches, to the Identity Services Engine (ISE), and to secure firewalls. This makes it a central component of “resilient infrastructure services.”
Live Protect is the tool that reconciles the “machine speed” of attackers with the stability of enterprise IT. It is the “seatbelt” that protects you while you service the vehicle (the infrastructure) in the workshop (during patching). However, it is not a substitute for maintenance itself.
“Cisco not only provides the protection but also helps to remove it once it is no longer needed.”
Two Frontier partners at the same time (Glasswing + Daybreak)
Why does Cisco work with both Anthropic and OpenAI? What does one offer that the other does not? Should customers conclude that a multi-model approach makes sense for them as well?
Every AI model has “blind spots.” Anthropic (Mythos) and OpenAI (GPT-5.5-Cyber) use different training data and have different logical focuses. By using both models, we achieve broader coverage. For customers, this means: When securing your infrastructure, do not rely on a single AI source, but rather on an ecosystem that validates different approaches.
Do you want to learn more about Cisco & Project Glasswing?
If you are interested or would like to know more, contact us to schedule a meeting with our experts: sales@nts.eu.