Nikhil Handigol is Co-Founder and Chief AI Officer, Forward Networks, building the foundation for autonomous networking.
The most dangerous moment on an enterprise network is not when something noticeably breaks, but when a change works exactly as designed yet opens a door that nobody meant to open.
I think about this constantly, because it sits at the heart of a divide that most organizations have never fully reconciled. Network operations teams are measured on connectivity, performance and uptime. Security operations teams are measured on containment, exposure and risk.
Even when both do their jobs well, they typically work from different tools, different network insights and different definitions of success. That divide becomes most perilous during the change window.
Why One Change Equals Two Truths
Consider a routine change in your organization, such as when a team migrates an internal application from an aging server rack to a modern data center. In this scenario, let’s say the application comes up as expected, stays reachable from everywhere it needs to be reachable and every functional check passes.
From the network team’s point of view, the change is a clean win, since the routing and application both worked.
What that same change can also do, without anyone noticing, is expose the internal application to the internet, opening a path from the outside world to a system that was never intended to be reachable. Since there was no disruption to notice, it’s likely no one will detect this change in security posture until there’s an incident.
This is why the disconnect between the two teams matters. In spite of the best efforts to safely conduct changes, planning, change advisory board (CAB) reviews and even lab testing, there is no way to definitively determine the full impact of a change on network behavior and security posture without pushing it live. In other words, the production network is the testbed.
The Exposures That Never Trigger An Alarm
The exposures that worry me most are changes that seem to deliver the intended outcome but trigger an incident in the future.
For instance, leaving a decommissioned legacy system accessible from the internet can leave an unknown security hole. This is what security researchers suspect happened when data from data broker DemandScience appeared for sale on a hacking forum, according to SiliconANGLE.
Broadly speaking, it’s irrelevant if the checklist says the system is gone when the network says otherwise. Nobody will have a way to reconcile the two until the data is already circulating.
Attackers also now use AI to turn an unknown weakness into a working exploit at machine speed. An exposure that used to be survivable for days can now be found and used in minutes. When that is the threat model, a silent exposure that lingers even briefly in production is not an acceptable cost of doing business.
Security and network teams must have deterministic knowledge of the impact of their changes before they are implemented. Anything less is an untenable level of risk.
Close the Gap Before the Window Opens
My company operates in the network digital twin space, building a mathematically accurate model of an organization’s production network that can be used to understand the impact of a change before it is made. That experience naturally informs my view. But the broader principles are independent of any particular vendor or technology.
We must focus on closing the gap between network intent and security reality before a change goes live, not after.
The first requirement is to shift network operations left. If networks are genuinely mission-critical infrastructure, we should take a lesson from other mission-critical industries: Do not make the production environment the place where you discover whether something works.
That starts organizationally. Network and security teams need shared criteria for what constitutes a safe change, common policies for connectivity, exposure and compliance and a clearly defined approval process. They also need to work from the same evidence about how the network behaves today and how that behavior will change after an update, rather than reconciling two different interpretations after the fact.
Technology should reinforce that operating model. That can include automated configuration analysis, policy and compliance checks, testing environments, emulation and simulation, change automation and network digital twins. The objective is the same: Move as much validation as possible ahead of production and replace assumptions with evidence.
Before any change reaches the live network, a team should be able to answer two questions with confidence.
First, does the change accomplish what we intended?
Second, does it alter the organization’s security posture in any way we did not explicitly approve?
This becomes even more important as AI moves from advising network operators to acting on their behalf. AI can dramatically accelerate operations, but without the right guardrails, it can also accelerate mistakes, unintended connectivity and security exposure at machine speed and scale.
An AI agent should therefore not be trusted simply because it can generate a plausible change. Any proposed action should be evaluated against policy and validated in a trusted model, simulation or equivalent pre-production environment before it is allowed to touch the live network.
Digital twins can play a role because they allow the consequences of a proposed change to be evaluated against the actual state of the network before execution. Whatever technology an organization chooses, the principle is the important part: AI should operate inside verifiable constraints.
Security posture should never be something you learn about after the fact. It should be something you prove before you commit. When network and security teams share the same evidence, validate outcomes before production and put enforceable guardrails around both human and AI-driven change, the change window stops being the most dangerous moment on the network and becomes the most controlled one.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?