Adversarial System Resilience: When Complexity Becomes a Vulnerability
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?
Authors
- Mikhail Gorelkin
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-14
- DOI
- https://doi.org/10.5281/zenodo.24228629
- Primary Topic
- Infrastructure Resilience and Vulnerability Analysis
- Type
- article
- Field-Weighted Citation Impact
- 0.00