Work

Adversarial System Resilience: Diagnosing the Hidden Risks of Protective Mechanisms

Year: 2026
Type: article
Abstract: This paper continues Adversarial System Resilience: When Complexity Becomes a Vulnerability (ASR; Complexity Lab, Zenodo, 2026; DOI: 10.5281/zenodo.24228630). ASR frames the problem: a vulnerability of a complex system can arise from the dynamics of the whole, and an adversary can exploit the system's own responses. Here that framing is applied to one class of mechanisms, protective mechanisms, and developed into a diagnostic procedure. Protective mechanisms (lockouts, limits, retries, autoscaling, emergency modes, recovery procedures) are part of the system, and when they activate they change its behavior. The paper examines the risks that arise from the interaction of such mechanisms when each of them works correctly and meets its own requirement. The paper contains an analysis of a well-known outage; seven patterns, each with a supporting case and an evidentiary status; a diagnostic procedure, the interaction map of one critical process, which is proposed and shown on a hypothetical example but has not yet been tested in an organization; and a synthetic example with a reproducible queueing model. In the model, a ten-second burst of load moves request verification from temporary overload into self-sustaining degradation; 2.5 seconds at five times the normal rate, about forty additional requests, is enough for the transition. The model also shows how an ordinary engineering decision shifts the boundary of autonomous recovery, one of the boundaries of the process's viability, without worsening current service metrics. Such a shift can be noticed in advance by analyzing load, verification capacity, and retry policy together. As in ASR, three claims are kept distinct: that an effect is possible, that it can be induced deliberately, and that its consequences can be exploited. The paper is addressed to information security leaders, architects, and reliability specialists, and describes a bounded diagnostic project for one process. The Python source code that reproduces the numerical results is provided, with a README, in the same Zenodo record.
Language: English
Open Access status: green
OpenAlex ID: W7234615891
FWCI:
Cites:
Cited by:
Related to: 10