top of page
Search

A Thousand Points of Failure: The Internet's Bad Month

kljgamer1
10 minutes ago
6 min read



Given the tsunami of outages that have occurred this month, I have been getting a plethora of questions regarding whether or not this is a cyber attack. As one might imagine, lots of folks in my profession have opinions on this issue...


...here's mine.


What Has Happened Lately


Since the widely publicized Microsoft 365, FAA, and Amtrak service disruptions, several additional significant incidents have occurred:

  • JPMorgan Chase (November 16): A technology failure disrupted online banking and transaction processing. Although customers of several other financial institutions reported problems, subsequent reporting suggests that at least some of those issues were downstream consequences of the Chase disruption rather than independent banking outages.

  • United Airlines (November 16): A technology failure affected systems used by flight crews, resulting in operational disruptions and flight delays.

  • National Security Agency (November 17): The NSA acknowledged degradation of internal systems, reportedly including its classified communications network. The agency attributed the disruption to scheduled maintenance and indicated that mission-critical capabilities remained operational.


These incidents join a growing list including GitHub, Cloudflare, Anthropic, and other major technology providers.


While these events span multiple industries, their proximity in time is noteworthy. The more important question, however, is whether they share an underlying cause or vulnerability.


What Are We Learning About the Causes?

New technical disclosures are beginning to provide a clearer picture, particularly regarding the Microsoft and FAA incidents.


  • Microsoft: Microsoft's preliminary analysis describes a software defect involving contention for shared computing resources. The defect had apparently existed for months without causing a major problem. However, a combination of increased traffic, reduced available capacity, and automatic traffic redistribution triggered a cascading failure. In simpler terms, a weakness that had been dormant under normal operating conditions became catastrophic when several otherwise manageable conditions occurred simultaneously. Importantly, the mechanisms intended to maintain service availability actually contributed to propagating the failure.


  • FAA: The FAA's investigation identified instability within legacy telecommunications infrastructure associated with unusually high volumes of connection attempts and ongoing router replacement activities. Isolating affected network components stabilized the environment. While technically different from Microsoft's incident, the underlying pattern is similar: existing infrastructure became unstable when exposed to conditions it was not sufficiently resilient to handle.


  • JPMorgan Chase and NSA: Initial explanations point toward internal software, infrastructure, and maintenance-related problems rather than malicious activity.


Taken together, these findings suggest that at least some of the recent incidents reflect weaknesses in technology resilience, particularly where complex dependencies, legacy infrastructure, and automated recovery mechanisms interact. That does not establish a common cause, but it does identify a common class of failure.


An Emerging Variable: AI-Enabled Offensive Cybersecurity Tools


There is another development that I believe deserves attention.


On November 13, the FBI issued an advisory regarding the weaponization of ARTEX, an AI-enabled offensive cybersecurity framework. Other security researchers, including Google/Mandiant, have documented the growing use of autonomous tools capable of conducting reconnaissance, vulnerability discovery, exploitation, and other offensive activities at machine speed. These capabilities represent a meaningful evolution in the threat landscape. Historically, sophisticated attacks required substantial human effort to identify vulnerabilities, test systems, and adapt techniques. AI-enabled offensive frameworks can increasingly perform portions of those activities autonomously, in parallel, and at considerably greater speed.


Several colleagues of mine who operate managed security services organizations have reported dramatic increases in serious security incidents over the past several months. One described an increase from approximately seven major incidents every two weeks to more than 30 per week. These are observations from individual providers rather than independently validated industry-wide statistics, but they are consistent with a threat environment experiencing significant acceleration.


What makes this relevant to the recent outages is that an attacker does not necessarily need to compromise a system to adversely affect its availability. Consider a scenario in which numerous autonomous tools are simultaneously probing applications, testing authentication mechanisms, examining APIs, and repeatedly attempting transactions. Even when these activities are blocked, the underlying systems must expend resources processing requests, evaluating security controls, recording activity, and responding. Under the wrong conditions, this activity could increase latency, exhaust computing resources, or trigger automated recovery mechanisms. Those mechanisms can sometimes amplify rather than contain the original problem. Put differently, a technology environment designed to withstand conventional attack volumes might experience unexpected instability when subjected to sustained, adaptive, machine-speed activity. This is a technically plausible scenario, particularly for older systems or environments with complex dependencies.


To be clear, there is no evidence at present that ARTEX or similar tools caused the Microsoft, FAA, banking, or other outages. The published explanations for several incidents point strongly toward internal technology failures. Nevertheless, I believe the potential interaction between rapidly increasing offensive automation and existing technology fragility deserves serious consideration.


Are These Events Connected?


At present, I see three possible forms of linkage, each with different implications.


1. Direct downstream dependencies. This is already demonstrable in several cases. Microsoft's outage affected numerous third-party applications and business processes dependent upon its services. Similarly, problems at one financial institution can affect transactions involving other banks even when those institutions' own infrastructure remains operational. This is the clearest and best-supported form of linkage.


2. Common architectural and operational weaknesses. This is, in my assessment, the most likely broader explanation. Organizations across industries have built increasingly complex environments involving interconnected applications, cloud services, legacy infrastructure, and automated failover capabilities. These architectures are highly capable but can also contain hidden dependencies and failure modes that become apparent only under unusual operating conditions. The Microsoft and FAA incidents illustrate how seemingly manageable technical problems can escalate into major disruptions. In this scenario, the events are connected not through a common attacker or shared infrastructure, but through similar vulnerabilities in how modern technology environments are designed, operated, and recovered.


3. Increasing external stress from autonomous offensive activity. This is the most speculative of the three possibilities, but potentially the most interesting. As AI-enabled offensive tools become more prevalent, they may create a sustained increase in background security-related traffic and processing demands. That additional stress could expose latent technology weaknesses or contribute to failures that would not otherwise occur. This would not require a coordinated campaign, a nation-state adversary, or even an explicit intention to disrupt services. It would simply require an increasingly hostile operating environment interacting with systems whose resilience assumptions may no longer be adequate. Again, there is insufficient evidence to conclude that this explains the current cluster of outages. It is a hypothesis, not a finding.


My Current Assessment


Based on the information available today, my assessment is as follows:

  • High confidence: Several major outages have identifiable technical explanations, and there is no established evidence linking them to a coordinated malicious campaign.

  • High confidence: Shared technology dependencies can propagate failures across organizations and industries, making a single incident appear to be numerous independent outages.

  • Moderate confidence: Common architectural weaknesses, insufficient capacity margins, and limitations in recovery mechanisms are significant contributors to the broader pattern.

  • Low confidence: AI-enabled offensive activity materially contributed to the specific outages experienced over the past several days. This remains plausible but unsubstantiated.

  • Very low confidence: The incidents represent a single coordinated cyberattack.


I would emphasize that these are analytical judgments based on currently available information, not conclusions drawn from comprehensive forensic access to the affected organizations.


What Does This Mean for Tech-Driven Organizations?


There are several practical implications.


First, organizations need to continue evaluating the resilience of our critical business services against failures originating outside our direct control. Our universal dependencies on, telecommunications providers, cloud platforms, identity services, and payment ecosystem participants make this particularly important.


Second, you should examine whether your resilience assumptions adequately account for multiple simultaneous stressors. A system that performs well under peak legitimate transaction volume may behave differently when that demand coincides with elevated malicious traffic, a degraded third-party dependency, or partial infrastructure failure.


Third, you should increasingly evaluate cybersecurity not simply in terms of whether an attacker can penetrate our defenses, but also whether sustained adversarial activity can materially degrade the performance or availability of critical business functions—even when the attacks themselves are unsuccessful.


Finally, you need to ensure that your monitoring and intelligence capabilities can distinguish malicious activity, normal operating demand, and cascading technical failures. That distinction will become increasingly important as AI-driven offensive capabilities mature.


A Final Thought

I believe the most important takeaway from this situation is that we may be entering a period in which the speed and scale of AI-enabled offensive activity are increasing faster than the resilience of the technology environments being targeted. That does not mean the recent outages were cyberattacks. In several cases, the evidence suggests otherwise. Still, it does raise a broader question: Are our technology environments sufficiently resilient for an operating environment in which automated adversarial activity is continuous, adaptive, and increasingly conducted at machine speed? That is a question worth asking regardless of whether these particular incidents ultimately prove connected.


My two cents...hope it helps!

 
 
 

Comments


  • LinkedIn Social Icon

© 2023 by Kim L. Jones.  All rights reserved.  Proudly created with Wix.com

Success! Message received.

bottom of page