---
title: Why Operational Resilience Demands More Than Strong Cybersecurity
slug: operational-resilience-cybersecurity
url: /topics/articles/operational-resilience-cybersecurity/index.md
publishedAt: "2026-11-17T13:51:35"
updatedAt: "2026-11-17T13:51:35"
excerpt: "Two widely discussed disruptions in November 2026, one affecting Microsoft 365 services and another affecting U.S. aviation operations, have renewed a familiar question: Was this a cyberattack? That is a legitimate question, but it is not a conclusion. At the time of this article, publicly discussed explanations remain preliminary and should not be confused with [&hellip;]"
authors:
  - Howard Penrose
---

# Why Operational Resilience Demands More Than Strong Cybersecurity

By [Howard Penrose](/authors/howard-penrose/index.md)

Two widely discussed disruptions in November 2026, one affecting Microsoft 365 services and another affecting U.S. aviation operations, have renewed a familiar question: Was this a cyberattack? That is a legitimate question, but it is not a conclusion. At the time of this article, publicly discussed explanations remain preliminary and should not be confused with completed root cause failure analyses (RCFA). The incidents should also not be treated as connected without evidence.

The more important engineering question is broader: How can an authorized change, a legacy component, an automated response, or a management decision produce consequences similar to those a malicious actor might seek? We do not have to prove hostile intent to examine the vulnerabilities that allow failures to propagate.

## A failure mechanism is not a root cause

In reliability engineering, identifying the component that stopped functioning is the beginning of an investigation, not the end. If a communications switch becomes overloaded, the investigator must still establish why the traffic increased, why protective functions did not contain it, whether the triggering condition was anticipated, and whether testing and change management were adequate. The same discipline applies to software and cloud services.

The November incidents have prompted theories ranging from coordinated attacks to artificial-intelligence-driven changes. Such theories should be recorded as hypotheses only when they can be expressed in testable terms. Their popularity, proximity in time, or political appeal does not constitute evidence. A complete investigation must distinguish the initiating event, the physical or digital failure mechanism, contributing organizational conditions, and the consequence.

## The vulnerability may be a decision

For decades, industrial organizations have treated cybersecurity primarily as the defense of systems against unauthorized access. That remains essential. However, a fully authorized action can also create a critical vulnerability. A rushed network upgrade, an untested interface between old and new equipment, or a policy that rewards deployment speed over verification can defeat defenses without an attacker ever entering the network.

I have seen an industrial facility lose part of its operations when a legacy network switch interacted poorly with a newer monitoring device. Protective configuration and the response to unexpected network traffic mattered. Fortunately, the facility’s segmentation and redundancy helped limit the effect. The lesson was not that modern devices are dangerous; it was that integration assumptions must be tested under actual operating conditions.

![Figure 1. Illustrative causal pathway. Each arrow is a question for investigation, not a presumption of causality.](https://cms.reliamag.com/wp-content/uploads/2026/11/figure1-1.png)
*Figure 1. Illustrative causal pathway. Each arrow is a question for investigation, not a presumption of causality.*

## Red team the organization, not only the firewall

Red teams traditionally adopt an adversarial perspective to identify security weaknesses, while blue teams emphasize defense and response. The concept can be extended to examine organizational resilience: What happens if a legitimate change is wrong? What if a third-party service disappears for 24 hours? What if the people who understood the legacy equipment have retired? Who can stop a deployment, and who has the authority to reverse it?

This proposal is consistent with, but broader than, established cybersecurity governance. NIST Cybersecurity Framework 2.0 introduced an explicit Govern function and increased attention to supply-chain risk [1]. It does not require every staffing or management policy to undergo red teaming. Applying adversarial exercises to digital transformation, staffing, procurement, change approval, and recovery is an additional practice that industrial organizations can choose to implement.

![Figure 2. Proposed exercise model. Technical teams and management share accountability for the outcome.](https://cms.reliamag.com/wp-content/uploads/2026/11/figure2-1.png)
*Figure 2. Proposed exercise model. Technical teams and management share accountability for the outcome.*

## Artificial intelligence: accelerate analysis, preserve accountability

Artificial intelligence and machine learning are not new to industrial analysis. Pattern recognition, anomaly detection, and automated comparisons have supported condition monitoring for many years. The recent development is the speed and accessibility of general-purpose systems that can draft code, interpret documents, propose actions, and interact with enterprise applications.

These tools can be valuable during RCFA. A team can ask an independent system to enumerate overlooked hypotheses, find conflicting evidence, compare prior failures, or challenge an assumption that everyone in the room accepts. In an RCFA investigation an apparently unlikely explanation may turn out to be correct. That is why brainstorming must preserve ideas until evidence confirms or rejects them.

> A separate model is not automatically an independent witness.

However, a model can produce plausible errors, inherit biases, misinterpret operating context, and overstate confidence. A separate model is not automatically an independent witness. Evidence still requires traceability to measurements, records, testing, and qualified human review. Most importantly, an AI tool that recommends a change should not automatically have authority to execute that change in a safety-critical system.

## The lessons of CrowdStrike and aviation modernization

The July 2024 CrowdStrike disruption illustrates how an authorized software update can create widespread consequences. Microsoft estimated that approximately 8.5 million Windows devices were affected [2]. The significance is not that every update is hazardous; it is that concentrated dependencies can transform a small percentage of affected devices into a disproportionate operational problem.

For the FAA, the long-term infrastructure concern is documented independently of any single November event. The Government Accountability Office reported in 2024 that 51 of 138 air traffic control systems were unsustainable and 54 were potentially unsustainable [3]. In September 2026, GAO warned that accelerated modernization required stronger cost and schedule planning, and described integration risks involving old and new voice-switch systems [4]. These findings support the need for disciplined modernization and not an explanation of the November failure, which is still under investigation.

## Third-party access is an operational dependency

Another lesson comes from website and industrial technology incidents: a trusted supplier, backup utility, remote support service, or monitoring platform may provide an access path into systems that the operator believes are protected. A vendor’s monitoring service is not a substitute for an owner’s independent controls. Organizations should know what privileges suppliers hold, whether credentials can be revoked promptly, and whether operations can continue when a vendor becomes unavailable or compromised.

For industrial condition monitoring, a practical design choice is to separate observation from control. Diagnostic systems can produce information for technicians and engineers without automatically issuing commands to production equipment. Where integration with a computerized maintenance management system or control platform is desired, the owner should define the trust boundary, validate inputs, and retain decision authority.

## A practical resilience exercise

A meaningful exercise starts with a credible failure rather than a fictional supervillain. For example, assume that an authorized network change causes an unexpected flood of traffic while the organization’s primary collaboration platform is unavailable. Add a staffing constraint and a third-party support delay. The red team asks what breaks; the blue team demonstrates detection and containment; operations demonstrates safe continuation; management demonstrates decision authority and communications.

Measure time to detect, time to isolate, ability to operate safely in a degraded mode, restoration time, data integrity, and whether the organization can contact customers without its usual cloud platform. Repeat the exercise with the human decision-makers removed or unavailable. Redundancy that depends on the same identity provider, network path, supplier, or approval chain may not be independent redundancy.

## Conclusion: test the decisions before they become vulnerabilities

Digital transformation should improve the reliability and capability of an organization. It should not simply increase the rate at which a poor decision can spread. There is a role for AI, automation, remote monitoring, cloud services, and modernized infrastructure. There is also a role for experienced personnel who understand consequences that cannot be inferred from a dashboard.

> The most dangerous single point of failure may not be a piece of equipment. It may be an assumption nobody was allowed to question.

The objective is not to resist technology. It is to demand evidence that technology has improved the system, including its ability to fail safely and recover. Red team the technology, certainly. Then red team the policies, incentives, staffing assumptions, supplier dependencies, and management decisions surrounding it. The most dangerous single point of failure may not be a piece of equipment. It may be an assumption nobody was allowed to question.

## References and editorial verification notes

[1] NIST, “NIST Releases Version 2.0 of Landmark Cybersecurity Framework,” Feb. 26, 2024. [https://www.nist.gov/news-events/news/2024/02/nist-releases-version-20-landmark-cybersecurity-framework](https://www.nist.gov/news-events/news/2024/02/nist-releases-version-20-landmark-cybersecurity-framework)

[2] Microsoft, “Helping our customers through the CrowdStrike outage,” July 20, 2024. [https://blogs.microsoft.com/blog/2024/07/20/helping-our-customers-through-the-crowdstrike-outage/](https://blogs.microsoft.com/blog/2024/07/20/helping-our-customers-through-the-crowdstrike-outage/)

[3] U.S. GAO, GAO-24-107001, “Air Traffic Control: FAA Actions Are Urgently Needed to Modernize Aging Systems,” Sept. 23, 2024. [https://www.gao.gov/products/gao-24-107001](https://www.gao.gov/products/gao-24-107001)

[4] U.S. GAO, GAO-26-107992, “Air Traffic Control Systems: Ambitious New Modernization Effort Needs to Improve Cost and Schedule Planning,” Sept. 14, 2026. [https://www.gao.gov/products/gao-26-107992](https://www.gao.gov/products/gao-26-107992)

