Photo by NASA on Unsplash
THE UNCERTAINTY DEPARTMENT
Cognitive Security Essays
In 1989, the East German Stasi employed 91,000 full-time officers and 173,000 informants to monitor and manipulate the behavior of 16 million citizens. The operation cost billions of dollars and required the organizational capacity of a small state.
Today, a behavioral modification campaign targeting millions of Americans can be designed and executed by a small team for under $100,000. This is not a projection. It is a present condition.
The technology that made this possible did not arrive as a single invention. It accumulated through the convergence of several independently developed capabilities: algorithmic content distribution systems optimizable for behavioral outcomes; synthetic media generation at industrial scale; micro-targeting infrastructure derived from commercial advertising; and social network automation tools that produce artificial social proof at low marginal cost. None of these capabilities was developed for use in behavioral modification campaigns. Each was developed for commercial purposes. Their convergence created a behavioral modification capability that now operates as a commodity.
The strategic significance of this shift is not that behavioral modification is now cheaper. It is that the barrier to entry has dropped below the threshold of state-level resources. A foreign intelligence service can deploy these capabilities. A domestic political operation can deploy them. A commercial interest seeking regulatory outcomes can deploy them. A network of private individuals coordinating around a shared objective can deploy them. The Stasi needed a state. The contemporary equivalent needs a budget and a platform account.
The historical category of propaganda describes influence operations that work through the content of their message — by presenting arguments, narratives, or information that the target evaluates and, if persuaded, accepts. The operator’s goal is to change what the target believes by changing what the target thinks about.
Contemporary behavioral modification campaigns work differently. Their primary mechanism is not the content of any specific message but the structure of the informational environment within which the target forms beliefs. The operator does not need to persuade the target to believe X. The operator needs to construct an environment in which X is what the target arrives at through their own apparently normal process of information evaluation.
This distinction is consequential for three reasons.
First, it defeats standard defenses. Media literacy education, fact-checking, critical thinking curricula — these tools are designed to help people evaluate specific claims on their merits. They do not address the prior question of whether the informational environment in which the person encounters those claims has been engineered to produce a specific outcome.
Second, it is invisible by design. A person who has been exposed to overt propaganda knows they have encountered a message designed to persuade them. A person whose informational environment has been algorithmically structured to produce a specific belief formation outcome experiences that outcome as their own independently formed view. The operation’s effectiveness depends on its invisibility. The more successful it is, the less the target knows it is occurring.
Third, it scales to populations without proportional cost increase. Persuading one person through direct argument costs roughly the same per person as persuading a million people through the same argument. Structuring an informational environment for a million people costs only marginally more than structuring it for one thousand, because the infrastructure cost is fixed and the marginal delivery cost approaches zero. The capability is structurally suited to mass application.
Climate denial is not the most important contemporary application of behavioral modification technology. It is the most important demonstration of it.
The scale of what was achieved is the relevant fact. A multi-decade behavioral modification campaign produced a condition in which a substantial portion of a highly educated democratic population maintains beliefs about a specific factual question that are directly contradicted by the most extensively documented scientific consensus in human history, by the direct observation of changing physical conditions, and by the testimony of the world’s major scientific institutions. This belief state has been maintained, across generational cohort turnover, in the face of continuous counter-messaging from governments, educational institutions, and scientific organizations.
Behavioral modification campaigns have achieved this against evidence that is:
If behavioral modification can achieve this against those conditions, the capabilities available for domains where the evidence base is less robust, less directly observable, and less actively contested are substantially greater.
The operational lesson is not about climate. It is about demonstrated capability. The question that follows is: what is being done with this demonstrated capability right now, in domains where the conditions for behavioral modification are more favorable than in climate?
The technical literature on behavioral modification through algorithmic platforms uses various terms — filter bubbles, engagement optimization, recommendation system manipulation — that describe symptoms rather than mechanisms. The mechanism is:
Step 1: Profiling. The target’s behavioral patterns — content engagement, network structure, response to various emotional framings, political and commercial behavior — are mapped through continuous passive observation. This profile is not simply a description of what the target believes. It is a model of how the target’s belief-formation processes respond to different informational inputs.
Step 2: Environment Structuring. The target’s information environment is structured according to the profile — not to expose them to specific content, but to ensure that the content they encounter through their normal information consumption behavior is distributed in ways that produce the desired belief formation outcome. The target does not receive messages; they receive a curated version of reality that is indistinguishable, from inside it, from the ordinary uncurated version.
Step 3: Social Proof Engineering. Simultaneously, the target’s social environment — the apparent beliefs and behaviors of their peers and reference groups — is managed to create the impression that the desired belief formation outcome is normatively appropriate. Social proof is among the most powerful factors in belief formation; engineering it is proportionally consequential.
Step 4: Resistance Inoculation. Advanced campaigns include a resistance management component: exposure to weak forms of counter-arguments, in contexts that prime dismissal, that inoculate the target against stronger counter-arguments they may subsequently encounter. A target who has been primed to find certain evidence sources untrustworthy will discount evidence from those sources regardless of its quality.
None of these steps requires the target’s awareness. The operation is most effective when the target is least aware of it. The target who understands that their information environment has been structured to produce a specific belief formation outcome is a target who can potentially resist the structuring. The target who believes they are forming their own views through normal independent information evaluation is a target who is defenseless.
The gap between the demonstrated capability of behavioral modification campaigns and the institutional response to that capability is not explicable through ignorance. The relevant actors — regulatory agencies, legislative bodies, intelligence services — are aware of the capability. The gap is explicable through institutional architecture.
Regulatory frameworks for information and communications were designed for a world in which information influence operated through the content of discrete messages that could be evaluated and potentially regulated. The FTC can act against specific false claims. The FEC can regulate specific political communications. Courts can adjudicate specific defamatory statements. None of these frameworks was designed to address the structural engineering of information environments that produces belief formation outcomes without any single actionable message.
The result is a systematic regulatory gap precisely where the capability is most consequential. Behavioral modification operations that produce the most significant effects — by operating at the environmental level rather than the message level — are the operations that fall most completely outside existing regulatory frameworks.
This is not an accidental gap. The institutional architecture of information regulation reflects the political economy of the industries it regulates, and the industries that most benefit from the regulatory gap are the industries with the most resources to invest in maintaining it. Legislative efforts to address algorithmic transparency, recommendation system accountability, or behavioral modification liability have faced consistent, well-funded opposition from precisely the parties whose business models depend on the gap’s existence.
The cognitive security framing — treating behavioral modification capability as a threat to be defended against — is analytically adequate but politically insufficient. The deeper stakes are constitutional.
Democratic self-governance presupposes that citizens form preferences through processes that are meaningfully their own — that the political will expressed through democratic institutions reflects something genuinely produced by the deliberating citizens, rather than an output engineered by actors whose interests may be systematically opposed to the citizens’ interests. The philosophical foundation of democratic legitimacy is that the governed are, in some meaningful sense, governing themselves.
Behavioral modification campaigns operating at scale on democratic populations do not merely influence the outcomes of democratic processes. They challenge the epistemic foundations on which those processes’ legitimacy depends. If a substantial portion of a democratic electorate is forming political preferences through a process that has been structurally managed by actors with private interests in those preferences, the question of whether those preferences are genuinely the electorate’s own — and therefore whether the democratic outcomes that follow from them are genuinely legitimate — is not a rhetorical question.
It is a question about the conditions under which democratic governance is possible at all.
The answer currently being given by the institutional architecture is: democratic governance is possible under any informational conditions, because the procedures of democracy — voting, representation, legislative process — are independent of the epistemic conditions under which citizens form the preferences those procedures aggregate. This answer is constitutionally adequate in the narrow sense and philosophically untenable in the broader sense. Procedures that aggregate preferences acquire their legitimacy from the quality of the preference-formation process. Engineering that process degrades the legitimacy that the procedures are supposed to express.
The demonstrable facts are these: the capability exists, is widely deployed, has produced measurable large-scale effects on democratic populations, falls substantially outside existing regulatory frameworks, and faces institutional responses that are systematically inadequate to its scale and sophistication.
The analytical inference is that the question is no longer whether behavioral modification campaigns are affecting democratic processes. The question is which democratic processes, in which domains, to what extent, by whom, and toward what ends.
These are questions that can be investigated empirically. The evidence base for behavioral modification campaigns is substantially richer than is typically assumed: platform behavioral data, advertising targeting records, attribution research from academic and intelligence community sources, and the accumulating record of specific campaigns studied in retrospect. The gap between available evidence and institutional response is not primarily an evidence gap. It is a will gap.
Closing that gap requires treating the manipulation of cognitive environments as a category of harm analogous to the manipulation of physical environments: something that the legal and regulatory architecture is obligated to address, not because it involves false statements that can be fact-checked, but because it produces real harms to individuals and institutions that have standing to seek remedy. The Crimes Against Consciousness framework developed elsewhere in this series provides one legislative vehicle for doing so. The cognitive infrastructure hardening program outlined in the accompanying policy brief provides an institutional vehicle. Neither is sufficient alone. Both are more adequate than the current architecture of nothing.
The population whose cognitive environment is being managed is, in an important sense, the subject of what is being done to it and the agent of any response. Whether it can be both simultaneously is the question that the next decade will answer.
THE UNCERTAINTY DEPARTMENT. This essay is part of the Cognitive Security series. See also: CSI-2025-PBRIEF-003 (Cognitive Infrastructure Hardening), OGE-2026-WP-014 (Non-Zero-Sum Narrative Contests).