Cruelty does not circulate online as a single act. It circulates as a process. This process has identifiable stages, and its persistence does not depend on the beliefs, intentions, or coordination of any one participant. It depends on how these stages couple together under conditions of low friction, delayed attribution, and displaced cost. What appears as chaos resolves, under examination, into a stable loop.
Conservation is defined operationally. Violence persists when propagation capacity plus replication capacity exceed enforceable suppression capacity under prevailing incentives. Where this inequality holds, circulation stabilizes regardless of moral orientation.
The first stage is production under low visibility. Extreme harm is generated in environments characterized by weak oversight, inconsistent enforcement, and low immediate consequence. These environments are not defined primarily by culture or ideology. They are defined by enforcement asymmetry. The cost of producing evidence of harm is lower than the cost of being identified as responsible for it. Documentation itself becomes cheap. The act may be ritualized, casual, opportunistic, or coercive, but structurally this variation is irrelevant. What matters is that production occurs where attribution is difficult and sanction is unlikely.
The second stage is extraction and reframing. Harm does not usually enter large networks directly from its point of origin. It is extracted by third parties (witnesses, activists, aggregators, journalists, or intermediaries) and reframed as exposure, warning, proof, or moral alarm. This reframing is often sincere. It is motivated by an attempt to stop the harm, force recognition, or mobilize response. Structurally, however, extraction is a transport operation. It moves material from a low-visibility environment into a high-visibility one. At this moment, the harm is no longer only an event. It becomes content.
The third stage is platform processing. At this stage, moral intention ceases to matter. Platforms do not route signals according to ethical orientation. They route signals according to salience, interaction density, and manageability. Outrage, denunciation, shock, and horror reliably generate engagement. Recommendation and ranking systems trained on engagement signals cannot distinguish condemnation from endorsement in any way that affects propagation. Both register as interaction. The system does not need to endorse harm for harm to circulate. It needs only to remain indifferent to valence while optimizing for continuity.
The fourth stage is amplification and aggregation. Once circulating at scale, the material detaches from its originating context. It is clipped, mirrored, subtitled, translated, and re-posted across networks. Each iteration reduces contextual specificity while increasing audience reach. This produces a qualitative shift. The harm is no longer a discrete occurrence but a persistent presence. At this point, the system acquires memory. Content survives individual removals because replication outpaces enforcement. Minor alterations are sufficient to evade detection. The signal persists even as instances are removed.
The fifth stage is adaptive response. Actors at the production stage observe the effects of circulation. They learn, implicitly or explicitly, what persists, what escalates, and what attracts reaction. Adaptation follows. This does not require a desire for notoriety or ideological alignment with platform incentives. It requires only observation under asymmetric cost. Escalation may occur not because attention is sought as such, but because the system rewards persistence of signal. The feedback loop closes without coordination.
Taken together, these stages form a closed circulation loop: production > extraction > amplification > adaptation > renewed production. No stage requires moral endorsement of cruelty. No stage requires explicit escalation intent. Each stage is locally rational under its constraints. The loop persists because nothing within it forces a discontinuity. Harm remains visible, discussed, condemned, and redistributed without becoming structurally expensive for the system that carries it.
What is absent from this loop is conversion. Recognition occurs. Condemnation occurs. Attention accumulates. But recognition does not reliably convert into enforceable consequence. That conversion requires additional conditions: clear attribution, jurisdictional authority, anticipatory intervention, and liability that lands inside the system rather than outside it. Where these conditions are fragmented or absent, recognition accumulates without resolution.
This explains a recurring pattern that otherwise appears perverse. Repeated exposure does not exhaust the system: it habituates the audience. Outrage does not destabilize circulation: it feeds it. Moderators absorb trauma. Investigators pursue attribution across fragmented jurisdictions. Activists expend labor amplifying alarm, a form of unpaid externalized work that stabilizes circulation without forcing internal cost. The harm remains present, but the cost of opposing it is externalized. The system continues to operate.
At this point, an intuitive explanation presents itself: that persistence indicates a claim of indifference or deviance among participants. This inference is high-adoption, but structurally misleading. The loop does not persist because participants lack a missing virtue. It persists because the system has no requirement to internalize the costs generated by harm circulation. Where cost does not land, persistence is rational.
This loop also clarifies why certain forms of harm appear disproportionately visible. Apparent concentration does not reliably indicate higher incidence. It indicates longer lifecycle, weaker sanction, lower attribution pressure, and greater compatibility with platform incentives. Harm survives where it can circulate cheaply. Elsewhere, equally severe harm is truncated before aggregation, creating the illusion of moral difference where there is primarily a governance difference.
At this stage, the reader may still believe that improved moderation, better activism, or stronger norms could interrupt the loop. This belief rests on an unexamined assumption: that recognition tends toward correction. The next section addresses that assumption directly. It does not deny the ethical appeal of recognition. It shows why, under specific structural conditions, recognition can stabilize rather than interrupt harm.
Information Failure Modes in Networked Violence
Elijah Morgan
© 2026 Elijah Morgan
This work is licensed under the Creative Commons Attribution 4.0 International License (CC BY 4.0).
License: https://creativecommons.org/licenses/by/4.0/
DOI: https://doi.org/10.5281/zenodo.19599231
Version 1.1
Published 2026-04-15