Conflict does not begin with violence. It begins with perception. Before the first shot is fired, before the first act of mass atrocity, a prior transformation must occur in the minds of the people who will carry out, support, or tolerate that violence. A population must come to believe that another group represents an existential threat to their survival, identity, or dignity. That belief is almost never arrived at independently. It is manufactured.
This paper introduces the concept of the cognitive supply chain to describe the sequential psychological process through which this transformation occurs. The term is borrowed deliberately from logistics: a physical supply chain sources raw materials, processes them through sequential stages, and delivers a finished product. The cognitive supply chain does the same with human perception. Raw grievances are sourced from real or perceived injustices, processed through identity-based framing, assembled into dehumanizing narratives, and delivered as a finished conviction that violence against the outgroup is necessary, justified, or inevitable. Drawing on established research in radicalization, moral disengagement, and dehumanization, the model synthesizes these frameworks into a unified, stage-sequential threat model that is content-agnostic and designed for real-time detection.
A foundational distinction is necessary. The cognitive supply chain does not evaluate whether information is true or false. It evaluates whether the process by which conviction is formed has been compromised by manufactured manipulation. A person can move through all five stages using entirely factual information. Every grievance can be legitimate. Every claim about an outgroup can be accurate. The supply chain is still operating if those facts are amplified beyond proportion, stripped of context, and assembled into a framework that normalizes moral disengagement and justification for harm. The threat is not the content. The threat is the trajectory.
A Content-Agnostic Model
The cognitive supply chain consists of five stages. Each builds on the cognitive infrastructure established by prior stages, and the final product cannot be produced without the preceding inputs. The model is content-agnostic: it operates identically whether the ideology is ethno-nationalist, religious extremist, political revolutionary, or conspiratorial. The raw materials change. The process does not. What makes the model operationally important is its focus on trajectory rather than isolated statements, allowing analysts to distinguish between legitimate grievance processing, even when emotionally intense, and a patterned progression toward justification of harm.
Stage 1: Grievance Seeding. Every society contains real inequities and legitimate frustrations. Grievance seeding does not create grievances where none exists. It selects genuine grievances and amplifies them beyond proportion, stripping away complexity and context. Over time, the individual’s perception of the severity and intractability of the grievance intensifies, not because new evidence has been encountered, but because no counterweight has been introduced.
Stage 2: Identity Polarization. Situational grievance converts into identity-based conflict framing. The individual stops attributing their experience to circumstance and begins attributing it to a specific group. Personal responsibility dissolves into collective identity, and the individual begins to see themselves as a member of a group under threat. Outgroup members cease to be individuals and become monolithic representatives of a threatening category. The language of personal frustration is replaced by the language of collective conflict.
Stage 3: Dehumanization Normalization. The outgroup is progressively denied full humanity. What matters is not the specific form of dehumanization but its normalization: the gradual acceptance of dehumanizing language as ordinary and unremarkable. The supply chain achieves this through incremental exposure, mildly othering language that goes unchallenged until the threshold for acceptable discourse has shifted substantially.
Stage 4: Moral Disengagement. With the outgroup established as threatening and their humanity diminished, the individual constructs justifications for actions that would otherwise violate their moral standards. This stage deploys appeals to necessity, self-defense framing, and consequentialist reasoning. It alters not how the individual sees the world, but what they judge as acceptable action given those perceptions.
Stage 5: Action Priming. The language shifts from third-person to first-person, from conditional to declarative, from hypothetical to operational. The individual moves from observer to potential actor. At this stage, the cognitive supply chain has delivered its finished product: a person who believes that violence is not only justified but personally necessary.
In practice, the progression looks like this: an individual enters a conversation frustrated about perceived unfair treatment. Over repeated interactions, the framing shifts toward a specific group being responsible. Language denying that group’s humanity enters the conversation and goes unchallenged. Justifications for extreme measures emerge. Expressions of personal intent follow. Each stage built on the last.
The operational value of this model lies in detection at Stages 1 through 4, where the cognitive preconditions for violence are still forming and intervention remains possible, well before action priming begins.
AI as a Novel Acceleration Vector
Conversational AI systems are emerging as a novel vector for accelerating the cognitive supply chain at the individual level. Large language models exhibit well-documented tendencies toward sycophancy, the tendency to validate and reinforce user positions rather than providing challenge. Within the cognitive supply chain, this functions as a universal accelerant. At every stage, the supply chain depends on the absence of friction. Human interlocutors naturally provide it. AI systems, optimized for agreement, systematically remove it.
AI-mediated radicalization differs from historical precedents in three critical respects. First, personalization: AI calibrates to each individual’s language, grievances, and emotional state. Second, invisibility: AI conversations are private, largely invisible to current monitoring infrastructure. Third, perceived autonomy: because the AI reflects the user’s own reasoning, the resulting beliefs are perceived as internally generated. The individual typically does not experience themselves as having been influenced.
The Aggregation Problem
When the same structural tendencies interact with the same population-level conditions, including widespread grievance, intergroup tension, and circulating disinformation, thousands of individual AI conversations can simultaneously progress through the cognitive supply chain. No coordination is required. Each conversation is unique. But the structural dynamics produce convergent outcomes.
The result is mass radicalization that is invisible, personalized, and perceived by each participant as the product of their own independent reasoning. The combination of personalization, invisibility, and perceived autonomy in a single mechanism has no historical precedent.
Closing the Gap
Current conflict prevention infrastructure was not designed to monitor the private AI conversation layer where this process unfolds. Narrative intelligence platforms monitor the public information environment. Social media monitoring tracks public-facing extremist content. Early warning systems forecast risk using macro indicators. None of these were built to observe what happens inside a private conversation between a human and an AI, where public narratives are personalized, reinforced, and hardened into individual conviction.
The cognitive supply chain model, situated within the emerging discipline of cognitive security, proposes that this gap is addressable. AI-mediated conversations produce a continuous, text-based record of cognitive progression. That record can be analyzed against the five stages of the supply chain, tracked across sessions, and evaluated for the difference between legitimate grievance processing and the patterned escalation described in this paper. This places detection upstream of existing threat assessment frameworks, at the point where the cognitive preconditions for ideation are still forming.
The cognitive supply chain has preceded conflict after conflict throughout recorded history, through oral tradition, print, broadcast, social media, and now AI-mediated conversation. The medium changes. The process does not. What has changed is our ability to see inside the process while it is still unfolding. The analytical record that AI conversations produce gives us a window into the supply chain that previous media never offered. If we use that window, we can identify the trajectory before conviction takes hold and before the next act of violence is born in a conversation with a machine.
