Systems that do not internalize the cost of harm will stabilize around its persistence.
This work proceeds from a single, unavoidable determination: the persistent circulation, escalation, and monetization of extreme violent media across contemporary digital platforms is not an anomaly, nor the product of isolated malicious actors, nor a failure of content moderation in the narrow sense. It is the predictable outcome of platform architectures whose design, ranking, recommendation, and monetization systems select for high-arousal content, irrespective of moral valence, and whose governance mechanisms have failed to mitigate the resulting systemic risks despite long-standing visibility of harm [DSA Art. 34(1), Art. 35(1)].
The prevailing analytical error has been to treat such material as a matter of individual deviance (“sadists,” “extremists,” or “bad users”) rather than as an emergent property of sociotechnical systems optimized for engagement. This misclassification has produced a regulatory blind spot. Where harm is framed as episodic or actor-specific, enforcement defaults to takedown, account removal, or case-by-case moderation. Yet the observed persistence, reconstitution, and mutation of the material across platforms demonstrates that these measures do not meaningfully reduce harm. Under the Digital Services Act, such persistence is not neutral evidence of difficulty: it is an indicator of unresolved systemic risk arising from the design and functioning of the services themselves [DSA Art. 34(2), Recital 81].
The DSA establishes a due-diligence regime that explicitly requires very large online platforms and search engines to identify, assess, and mitigate systemic risks stemming from the dissemination of illegal content, negative effects on fundamental rights, public health, civic discourse, and public security [DSA Art. 34(1)(a)–(d)]. Crucially, this obligation is forward-looking and structural. It is not satisfied by demonstrating intent to moderate, nor by reporting volume metrics alone, nor by asserting neutrality of infrastructure. Where a category of harm persists at scale, crosses borders, and reproduces through platform affordances, the regulatory question is not whether individual items violate policy, but whether the system has been adequately reconfigured to prevent foreseeable recurrence [DSA Art. 35(1)].
This work therefore treats cruelty not as a moral deviation but as a systems diagnostic. The repeated appearance of extreme torture media (often recontextualized, mirrored, condemned, or commodified) functions as a signal that platform feedback loops are selecting for content that produces shock, outrage, and high retention. Recommendation and ranking systems trained on engagement signals cannot distinguish between condemnation and endorsement: both register as interaction. As a result, attempts to oppose the material through amplification frequently intensify its reach, creating secondary and tertiary harms including normalization, recruitment, desensitization, and monetization [DSA Art. 27(1), Art. 38, Art. 35(1)(c)].
Under the DSA, such amplification dynamics are not incidental. Platforms are required to provide transparency regarding the main parameters of their recommender systems and to assess the systemic risks arising from those systems’ operation [DSA Art. 27, Art. 34(2)]. Where recommender systems are shown to privilege high-arousal or extreme content, mitigation measures must extend beyond disclosure to include meaningful adaptation of system design, ranking criteria, and monetization pathways [DSA Art. 35(1)]. Failure to do so, once risks are identified, constitutes a failure of due diligence rather than an unfortunate externality.
The harms addressed here are not confined to the victims depicted. They extend to viewers, communities, and institutions. Exposure pathways disproportionately affect minors and vulnerable users, implicating obligations related to child protection and dignity [DSA Art. 28, Art. 34(1)(b)]. The circulation of such material also intersects with criminal, extremist, and extortionary networks that exploit threshold participation (forced viewing, sharing, or production) to bind participants through culpability. These dynamics mirror those observed in terrorist recruitment and organized crime, where irreversible acts function as mechanisms of control. Under the DSA, the convergence of content dissemination, monetization, and organized harm elevates the issue from moderation failure to public-security risk [DSA Art. 34(1)(d)].
A further regulatory failure has been the reliance on linear governance models: notice-and-action systems, takedown queues, and reactive enforcement despite clear evidence that these mechanisms do not interrupt non-linear propagation. The DSA’s systemic-risk framework implies that purely reactive moderation is insufficient where foreseeable harms persist. Where audits or public evidence show continued escalation, persistence becomes strong evidence of insufficient mitigation and warrants supervisory scrutiny, and requires platforms to implement mitigation measures proportionate to the severity and likelihood of harm, including structural changes to service design [DSA Art. 35(1), Recital 81]. Independent audits are mandated to assess not only whether measures exist, but whether they are effective in reducing systemic risks [DSA Art. 42(4)].
It is therefore no longer defensible to argue that platforms merely host content or that responsibility rests solely with users. The DSA establishes platforms, particularly VLOPs and VLOSEs, as infrastructure operators with commensurate responsibilities [DSA Art. 1(1), Art. 34–35]. Knowledge of harm without proportionate mitigation triggers enforcement pathways, including supervisory measures, interim remedies, and penalties [DSA Art. 51–54, Art. 73–76]. In this context, inaction is a decision that externalizes risk onto users, communities, and the public sphere.
The cross-border nature of the phenomenon further undermines fragmented responses. The material circulates across jurisdictions, languages, and platforms, exploiting regulatory asymmetries. The DSA addresses this explicitly through the role of Digital Services Coordinators, enhanced Commission oversight for VLOPs, and mechanisms for cross-border cooperation [DSA Art. 56–60, Art. 61]. Where systemic risks are documented, regulators are empowered to act irrespective of national origin of content or users.
This document is therefore not an advocacy text, a moral appeal, or an investigative exposé. It functions as a risk-identification and governance input within the meaning of the DSA. It supports supervisory inquiry, audit scrutiny, and assessment of whether the mitigation duties in Arts. 34–35 have been adequately discharged. [DSA Art. 34(3), Art. 42]. The remaining variable is not evidence, but regulatory will.
The question posed by this work is consequently narrow and unavoidable: given established obligations under the Digital Services Act, and given persistent, documented systemic harms arising from platform design choices, will competent authorities exercise the powers already granted to them? The answer to that question will determine whether the current trajectory, where cruelty is selected, amplified, and monetized by default, continues as a tolerated externality or is finally treated as the systemic risk it demonstrably is.
Throughout this text, terms that resemble psychological or ethical concepts are used only as compression labels for observable system behavior. They do not function as explanations, motives, or interior states. Where a claim cannot be reduced to selection effects, affordances, or attribution dynamics, it is not operative. Psychology is not rejected as a local descriptor: it is rejected as a sufficient systems explanation. Furthermore, appeals to culture, morality, intent, preference, abdication, or good and bad faith do not operate as explanations here. These domains may appear as local descriptors or compression labels, but they do not constitute sufficient explanatory grounds for the persistence, scaling, or routing of harm within networked systems.
This work does not reproduce primary material, datasets, or case archives. The omission is deliberate. Exposure to such material functions as an amplification vector and introduces affective distortion into analysis.
Empirical claims within this text are therefore presented as observed structural patterns derived from direct investigation, cross-platform monitoring, and synthesis of publicly available reporting.
Where formal citation is absent, the claim is not offered as proof but as a falsifiable observation of system behavior. The burden is not to accept the claim, but to test whether the described structure produces the observed persistence.
This text is anti-quotable by design. Aphorisms that can circulate as content without forcing parameter change are excluded. Where a sentence appears compressible into a shareable moral statement, it has failed its function.
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