A new model for infrastructure security: How Cisco defends against AI threats

Each CISO is being requested some model of the identical query: are we prepared for AI-powered assaults? Dive into how Cisco is reshaping its personal community, not as a posture checked yearly, however as a steady working mannequin.

Most enterprises are managing dangers based mostly on a menace mannequin constructed for a unique period. You set a danger threshold. You targeted on the vulnerabilities above that line — those important sufficient to maintain you up at evening. Every part beneath it, you managed. That was an affordable tradeoff.

AI-powered cybersecurity instruments have modified the mannequin. They don’t simply speed up recognized exploits; they will discover and weaponize every part beneath your threshold, together with the vulnerabilities you determined weren’t pressing and the legacy gadgets you hadn’t gotten round to changing. The bar hasn’t simply moved. It’s been dropped. That realization is reshaping how we function and defend our personal community at Cisco, and we expect it ought to reshape how each enterprise thinks about cyber protection.

“The stuff we used to not fear about — that’s now precisely what we fear about. The bar has been dropped, and we should rethink the entire mannequin.”

What we’re up in opposition to

Cisco’s company community carries site visitors for thousands and thousands of gadgets, 1000’s of functions, and a fast-growing inhabitants of AI brokers. It’s a prime goal for a similar adversaries our merchandise are constructed to cease.

For years, we’ve got operated on the identical vulnerability-patching mannequin most enterprises nonetheless use right this moment: vulnerability disclosed, patch developed, change-window scheduled, guide approvals collected, repair deployed. That cycle — measured in weeks — made sense when adversaries wanted months to weaponize a newly disclosed flaw. That window is now hours, with the trajectory pointing to minutes, and no quantity of course of enchancment closes a spot that huge.

With new frontier AI fashions, conventional approaches to defending the community are now not enough. The identical capabilities that assist us discover and repair vulnerabilities quicker are additionally touchdown within the fingers of menace actors who can now scan, exploit, and weaponize weaknesses at machine velocity. This dynamic extends effectively past our personal code: our broader provider ecosystem is racing to patch vulnerabilities whereas adversaries leverage these identical fashions to find and exploit them, typically in parallel. The result’s a quickly compressing window between disclosure and exploitation, forcing us to evolve simply as rapidly.

Our workforces give attention to discovering and fixing vulnerabilities and use accredited, commercially accessible AI coding brokers ruled by contractual and technical controls to scan advanced merchandise with thousands and thousands of traces of code. This helps us floor vulnerabilities that people alone may miss.

How we’re responding: See it. Show it. Include it. Substitute it.

Operationally, knowledgeable by our work with Anthropic’s Challenge Glasswing and OpenAI’s Dawn, in addition to different frontier fashions, we’ve reorganized our inner protection round 4 pillars, prioritized from the skin in — beginning with the broader provider and menace panorama and dealing inward to our personal atmosphere.

On this mannequin, instruments and brokers don’t function as a guidelines however as a steady loop, reinforcing one another at machine velocity.

Actual-time visibility first. Visibility informs what we validate. Earlier than we may speed up something, we wanted a centralized, constantly up to date image of our full assault floor — each asset, id, service account, cloud entitlement, and API. Actual visibility isn’t simply an asset stock. It’s figuring out who owns every asset, how important the asset is, and precisely how unhealthy issues can get if it’s compromised. That’s the inspiration for each choice.
Steady publicity validation, not periodic evaluate. Validation informs the place we deploy runtime protections. AI-powered adversaries don’t prioritize by the Frequent Vulnerability Scoring System (CVSS) rating. They chain lower-severity vulnerabilities into working exploits quicker than any periodic evaluate cycle can catch. We stopped chasing vulnerability lists. This can enable us to simulate actual assaults at machine velocity to repair what’s really exploitable, not what’s theoretically dangerous. Assault path evaluation tells you what’s in danger; severity scores alone don’t.
Runtime safety as a bridge, not a vacation spot. Runtime telemetry feeds again into visibility. Runtime safety comprises threats whilst you repair the foundation trigger. It buys time till the precise repair is prepared. The objective is a manufacturing atmosphere resilient sufficient to maintain working safely even beneath partial compromise.
Modernization as a strategic safety crucial. Modernization retains the entire loop operating on infrastructure constructed for change. Our focus is on hardening the inspiration — retiring end-of-life methods, eliminating insecure legacy companies, and positioning our infrastructure for quicker patching and larger resilience. That trendy basis is what unlocks superior runtime defenses like Hypershield-class segmentation, Dwell Shield, and the eBPF-powered Tetragon agent, which delivers real-time vulnerability shielding with out reboots or binary modifications — capabilities that merely can’t run on legacy.

How we’re prioritizing: Outdoors in

Some of the concrete shifts we’ve made is how we sequence our response. When the scope of publicity is massive and you’ll’t do every part without delay, triage construction issues as a lot as technical functionality.

Our method: work from the skin in. Web-facing edges carry the best publicity danger and transfer quickest, in order that’s the place we’ve targeted patching velocity and shielding first. As we transfer towards the core, the tempo turns into extra deliberate — the boundaries there are amongst our most important. The segments separating our largest safety zones — the firewalls defending our most delicate belongings — get prioritized as a result of defending them limits lateral motion and comprises blast radius if one thing will get by way of.

From there, each choice runs by way of the identical risk-based logic: decide what’s most uncovered, most weak, and what’s the correct response — take away it from the community, section it, apply runtime safety, or speed up the patch. Finish-of-life and unsupported belongings get eradicated or remoted. Externally exploitable vulnerabilities get addressed first. Property that may’t be patched inside operational home windows get runtime-first safety whereas remediation proceeds.

The Greater Shift

All of this factors to one thing extra basic than a quicker patch cycle. The mannequin we’re constructing towards isn’t a hardened fortress. It’s an agile and adaptable system that may transfer constantly to a safer state with out taking a time-out to do it.

“The sport is at all times being able to redeploy new, safe applied sciences. This notion that I’ve bought to take a time-out and do patching work — that’s the sport of the previous.”

Because the trade is coming into a interval of intense infrastructure evolution, companies should adapt safety practices and operational fashions to construct and keep resiliency. Our participation in trusted initiatives like Challenge Glasswing and Dawn offers us with the deep insights essential to navigate this shift, yielding fast modifications in how we function. However we aren’t achieved. As we constantly mature our working mannequin, we are going to proceed to show each functionality internally — at scale and in manufacturing — sharing our learnings and finest practices that assist our clients evolve their very own safety operations.

The window to get forward of AI threats remains to be open. The organizations that construct this operational muscle will compound their benefit. Those who wait compound their danger.

“We don’t simply promote the community; we defend each minute of daily with the identical instruments we provide to our clients.”

Jason Lish is Senior Vice President, Chief Info Safety Officer at Cisco the place he offers strategic management and oversight for Cisco’s Info Safety capabilities, together with enterprise info safety, information safety, assault floor administration, and safety operations. He additionally oversees worth chain safety and the Safety and Belief Group’s mergers and acquisitions service.