DECISION INTEGRITY
Cognitive Decision Integrity (CDI)
Treating AI-enabled influence and cognitive warfare as cyber-physical systems that manipulate the human information-processing pipeline — and the integrity discipline that defends it.
This page documents adversary capabilities and influence patterns for threat understanding, detection and defense. Capability descriptions are analytical threat models, not instructions. Where a capability has not been publicly demonstrated at scale it is marked Hypothetical. The defensive objective is to preserve the decision-maker's integrity, not to operationalize pressure.
From propaganda to cyber-physical system
Cognitive Decision Integrity (CDI) is the assurance that a decision-maker is processing information accurately, without external manipulation, and within their cognitive bandwidth. Where cybersecurity protects data integrity, CDI protects the integrity of the human decision process. AI enables adversaries to attempt to degrade CDI at greater speed and scale than traditional influence.
The shift is conceptual: AI-enabled influence is better understood not as louder propaganda but as a cyber-physical system that targets specific layers of the human information-processing pipeline. The defensive response is an integrity discipline, not a counter-narrative.
1. The Cognitive Stack
Just as a computer has a hardware/software stack, human cognition can be modeled as a stack of layers. Each layer is a potential target; each has a characteristic failure mode that defense must prevent.
Illustrative analytical model, not an actual system. Each layer is a potential target of AI-enabled pressure; each failure mode is what defense must prevent.
Perception
FAIL: IllusionATTACKSynthetic media, deepfakes, synthetic audio, sensory overload.
CDI FAILUREIllusion — the target perceives a reality that is not there.
DEFENSEMedia provenance (C2PA / content credentials), synthetic-media verification, sensory triage.
Attention
FAIL: DistractionATTACKSalience manipulation, notification flooding, pattern-interrupt.
CDI FAILUREDistraction — the target misses critical data or fixates on noise.
DEFENSEAttention hygiene, signal triage, low-priority channel muting near decision windows.
Interpretation
FAIL: MisattributionATTACKFraming, priming, linguistic anchoring, context stripping.
CDI FAILUREMisattribution — the target assigns incorrect causality or intent.
DEFENSEContext verification, seeking alternative framings, separating fact from frame.
Memory
FAIL: DistortionATTACKRepeated reinforcement, source-monitoring errors, false association.
CDI FAILUREDistortion — the target recalls manipulated material as established fact.
DEFENSEDecision journaling, source logging, delayed reassessment of recalled claims.
Decision
FAIL: SuboptimizationATTACKChoice architecture, sunk-cost traps, risk reframing.
CDI FAILURESuboptimization — the target chooses an option that serves the attacker.
DEFENSEStructural defenses: cooling-off, second confirmation, documented reasoning, pre-committed criteria.
Research methodology note: cognitive-science studies of manipulation combine neuroimaging (fMRI/EEG) with behavioral-economics experiments to identify correlates of influence; AI does not change the ethics of this, only the throughput of micro-experiments.
2. CDI-Degrading Attack Vectors
These are structural manipulations of the decision environment, not mere falsehoods. Each is presented as a threat model with its CDI impact and the countermeasure that closes the gap.
Consensus Illusion
MECHANISMA synthetic majority — coordinated accounts express one view in a compressed window to trigger the social heuristic.
CDI IMPACTThe target short-circuits independent analysis in favor of herd signal.
DEFENSEVerify source independence, not content consistency; detect coordination anomalies.
Temporal Salience
MECHANISMInformation timed to cognitive-fatigue windows, with affective tagging carried into later inputs.
CDI IMPACTRisk assessment altered by stress association rather than evidence.
DEFENSEFatigue-aware decision windows; defer consequential choices off high-load periods.
Source Credibility Poisoning
MECHANISMSlow, deniable erosion of trust in a target's own sources via minor inconsistencies.
CDI IMPACTEpistemic paralysis — doubt of all sources, reliance on intuitive (System 1) judgment.
DEFENSESource authentication; trust calibration; do not generalize one flaw to all sources.
Alternative Reality Bubble
MECHANISMStatistically true but contextually false data, with counter-context suppressed.
CDI IMPACTFragmentation — decisions on incomplete data produce systematic bias.
DEFENSERequire full context and base rates; actively seek counter-data.
Cognitive Load Overload
MECHANISMMany complex, framework-incompatible issues injected simultaneously to exhaust switching capacity.
CDI IMPACTThe target ignores the most important issue because it costs the most effort.
DEFENSEPriority protocols; limit concurrent high-stakes decisions; escalate the hardest item first.
3. Adversary Capability Model
What an adversary would need to mount individualized CDI pressure, and what each capability implies for defense. These are capability descriptions, not build steps; several are Hypothetical at the individual scale.
Target Concern Modeling
CAPABILITYAn adversary builds a model of a target's concerns from public data and, where access is compromised, behavioral traces (metadata, queries). Used to select which pressure to apply.
DEFENSEData minimization; monitor for unauthorized access; assume public data is already modeled.
Digital Twin / Counterfactual Simulation
CAPABILITYA behavioral model of the target used to simulate "what happens if stimulus X is injected at time Y" and select the highest-probability path. Hypothetical at the individual scale.
DEFENSEReduce decision predictability; introduce structural variation; do not routinize high-value decisions.
Persuasive LLM Pipeline
CAPABILITYPersona-tuned, RAG-grounded content generation producing "true but framed" material rather than obvious falsehoods, with style transfer to mimic trusted voices.
DEFENSEProvenance checks; style-anomaly detection; inoculation training; treat fluency as a signal, not proof.
Omni-Channel Orchestrator
CAPABILITYCross-channel timing coordination to create a "surround" effect — the same narrative appearing independently across email, professional networks, media and social.
DEFENSECross-channel convergence detection (Cognitive SOC); treat rapid multi-source agreement as a coordination signal.
See also the dedicated Cognitive Twin page for the modeling capability and Advanced PCP Techniques for the influence-pattern set.
4. CDI-Degrading Techniques
Granular behavioral-economics effects used to break the integrity of the choice itself. Each is a known cognitive bias; defense is to recognize the pattern and re-engage System 2 reasoning.
Anchoring & Adjusting
METHODAn extreme reference point is set first, so the real figure appears reasonable by contrast.
DEFENSEIndependent baselines established before any figure is offered.
Decoy Effect
METHODA dominated option is added to make the target option look like the smart choice.
DEFENSEPre-commit criteria; evaluate each option against needs, not against each other.
Sunk-Cost Escalation
METHODPrior investment is highlighted to frame stopping as loss.
DEFENSEPre-committed exit criteria; evaluate forward, not backward, from the decision point.
Availability Heuristic
METHODA rare event is repeated until it feels common, inflating perceived probability.
DEFENSEBase-rate checking; ask for frequency, not vividness.
Frictionless Path
METHODCognitive friction is removed from the desired option so it becomes the easiest choice.
DEFENSETreat suspicious ease as a signal; add deliberate friction to consequential choices.
- ▸CDI extends data integrity to the integrity of the human decision process.
- ▸The Cognitive Stack names five layers and five failure modes (illusion, distraction, misattribution, distortion, suboptimization) that defense must prevent.
- ▸AI-enabled vectors (consensus illusion, timing, trust decay, context isolation, load overload) are structural manipulations of the decision environment, not just lies.
- ▸Individualized closed-loop CDI degradation is hypothetical; no public system has demonstrated it reliably at scale.
- +Behavioral-economics effects (anchoring, decoy, sunk cost, availability) are well-established and exploitable.
- +AI persuasion and synthetic-media generation are demonstrated capabilities.
- +Cross-channel coordination is detectable when patterns align.
- ?Whether individualized closed-loop CDI degradation works reliably in the wild.
- ?Long-term resistance effects of cognitive inoculation against AI-personalized pressure.
- ?Where the legal line falls for undetectable manipulation across jurisdictions.
5. Open research questions
- Q1Detection: Can AI-generated social proof be detected before it shifts a decision? Look for abnormal consistency and compressed-window convergence.
- Q2Resistance: Does cognitive inoculation — pre-exposure to weak forms of manipulation — build resilience to stronger pressure?
- Q3Ethics: Where is the line between influence and manipulation when the method is designed to be undetectable?
- Q4Scale: Can personalized pressure scale to many targets with individual models, and at what cost? No public system has demonstrated this reliably.
AI-enabled influence is not about shouting louder; it is about shaping the cognitive environment. Defending Cognitive Decision Integrity means treating the decision process as an integrity surface — verifying provenance at the perception layer, protecting attention, checking context at interpretation, logging sources for memory, and installing structural defenses at the decision layer. The aim is not to win a narrative argument but to keep the decision-maker processing information accurately and autonomously.
SOURCES & REFERENCES
- [1]Thinking, Fast and Slow — Daniel Kahneman (2011). Foundational source on System 1 / System 2 cognition, anchoring, availability and cognitive load exploited by the vectors above. Source ↗
- [2]A Behavioral Theory of Rational Choice — Herbert A. Simon (bounded rationality) (1955). Origin of bounded rationality and cognitive-bandwidth limits underpinning Cognitive Decision Integrity. Source ↗
- [3]Influence: The Psychology of Persuasion — Robert Cialdini (social proof) (1984). Source for the social-proof heuristic exploited by the Consensus Illusion vector. Source ↗
- [4]Cognitive Warfare — NATO ACT (Ongoing) Source ↗
- [5]Regulation (EU) 2024/1689 (AI Act), Art. 5 — EUR-Lex (2024). Prohibits certain AI practices using subliminal or purposefully manipulative / deceptive techniques under specified conditions. Source ↗
- [6]C2PA Content Credentials Specification — C2PA (2026) Source ↗
- [7]Personalized AI persuasion (Matz et al.) — Scientific Reports (2024). Experimental evidence that LLM personalization can increase persuasion in controlled settings. Source ↗