Work

Adversarial System Resilience: When Complexity Becomes a Vulnerability

Year: 2026
Type: article
Abstract: Modern complex systems are still understood, designed, and protected predominantly in terms of components, functions, and known dependencies. Yet their vulnerabilities can arise from the dynamics of the whole: interactions, feedback loops, regime-dependent behavior, and changes in the ability to observe, intervene, adapt, and recover. Under adversarial conditions, these dynamics can be exploited without compromising individual components or possessing a complete model of the system. This paper develops Adversarial System Resilience (ASR) as a problem framing rather than a validated security methodology. Three cases support different, deliberately separated claims: the 2021 Meta outage shows how a failure can also disable the means of recovery; the 2025 Spotify outage shows a failure regime sustained by interacting responses; and the experimental Yo-Yo attack shows deliberate exploitation of cloud autoscaling dynamics. The paper distinguishes three claims: that a dangerous regime exists, that it can be induced from outside, and that an adversary can exploit its consequences. It also introduces a working notion of system viability, centered on preserving critical functions together with meaningful options for further action. Finally, recent infrastructure incidents motivate a broader, still-open question: how can changes in the conditions of a complex system’s viability be detected before their consequences surface as a large-scale disruption?
Language: English
Open Access status: green
OpenAlex ID: W7230039934
FWCI:
Cites:
Cited by:
Related to: 10