WMD-Class Risk Without Explosives: Autonomous Coercion in Cyber Systems

This Work-in-Progress paper examines a class of emerging cyber systems whose catastrophic risk does not arise from physical destruction, denial of service, or kinetic effects, but from the capacity to impose persistent, scalable coercion without renewed human consent. Unlike traditional cyber weapons, which operate as discrete tools activated by identifiable actors, these systems exert influence continuously, shaping behavior, constraining options, and enforcing outcomes as an ambient condition of the digital environment. As a result, force is no longer episodic or exceptional, but infrastructural, echoing classic concerns about cybernetics, control, and social organization [1, 4]. We argue that existing cybersecurity, artificial intelligence safety, and cyber conflict frameworks inadequately address this risk because they remain grounded in action-centric models of harm. These models emphasize exploits, payloads, and events, while overlooking systems whose primary effect is the automation of authority itself, consistent with long-running arguments that technical and institutional artifacts can embody politics and governance [2, 3]. The paper characterizes the structural properties that give rise to autonomous coercion, distinguishes it from deterrence and defensive control, and analyzes catastrophic failure modes associated with loss of control, escalation through misinterpretation, erosion of consent, and irreversible embedding in critical digital infrastructure [6, 7]. Importantly, we show that such risks remain acute even when systems are designed with benevolent intentions such as stability, safety, or harm reduction, as humanitarian framing can accelerate normalization and lower deployment thresholds [8]. We intentionally avoid system design, implementation details, simulations, or case studies, and instead focus on risk recognition before operational capability exists. The paper proposes bright-line boundaries, termination conditions, and governance considerations for the Science of Security community, arguing that early restraint and non-proliferation principles represent the last low-cost intervention point [5, 12]. By naming autonomous coercion as a distinct WMD-class risk without explosives, this work invites interdisciplinary scrutiny and community feedback before continuous machine-mediated enforcement becomes an accepted feature of the digital landscape.