Hallucinated Agents: A Cognitive Mechanism for Resolving Ambiguous Threats

Why do people freeze, misinterpret ambiguous cues, or make rapid, irrational choices during acute stress response? Why does panic erupt from seemingly nowhere, escalating from mild uncertainty to full‑blown panic attacks? And why do humans sometimes “rationalize” fear by inventing external threats or mislabeling internal sensations? This working paper proposes that these phenomena reflect not cognitive errors, but the operation of an evolved threat‑management system designed to prevent catastrophic indecision under conditions of high arousal and limited information. When fear, urgency, and ambiguity converge quickly, the defensive system must avoid prolonged unresolved states. High‑arousal ambiguity is metabolically costly and physiologically hazardous: sustained catecholamine surges, autonomic dysregulation, reduced HRV, and—at extremes—arrhythmia or stress‑induced cardiomyopathy (Takotsubo) can emerge from prolonged freeze or overload. To maintain bounded time‑to‑decision under asymmetric survival costs (missing a real threat is far worse than reacting to a false one), the system enforces a dual‑threshold stop rule embedded in the defensive cascade. When external ambiguity persists beyond person‑specific thresholds, the system inserts a Minimal Viable Agent (MVA)—a computationally minimal inferred adversary (with implied intent, location, trajectory)—which re‑anchors cognition to familiar threat vectors and restores actionable policy (orient, avoid, flee, inspect). This mechanism explains why ambiguous shadows, noises, or movements can trigger rapid, involuntary threat completion, functioning as an adaptive off‑ramp rather than a malfunction. For internal ambiguity (e.g., panic loops triggered by catastrophic misinterpretation of benign bodily signals as threats—heart attack, loss of control, “going insane”), the model proposes a parallel primitive: the Internal Minimal Viable Cause (IMVC). Here, the insertion is a benign reinterpretation (“this is adrenaline,” “this will pass”) that collapses catastrophic interoception and restores regulatory control. Unlike external MVA insertion (reflexive and pre‑deliberative), IMVC often requires more effortful re‑anchoring—a key reason panic attacks feel more acute and harder to terminate without intervention. The framework unifies disparate literatures under a single dual‑threshold controller: freeze/fight neural dynamics (amygdala–PAG–frontal transitions), threat appraisal, error‑management biases, embodied predictive processing, catastrophic misinterpretation models of panic, intolerance of uncertainty, and compensatory control. We formalize the controller with state variables for fear intensity, urgency, and ambiguity; define dual termination thresholds (τ* for external, λ* for internal); derive a composite risk function R(t); and provide pseudocode for the off‑ramp mechanism. Testable predictions span sensory modalities, cross‑modal congruence, arousal modulation, individual differences (Type M vs. Type P), and intolerance‑of‑uncertainty scaling. Appendix B outlines VR‑ready validation paradigms: low‑light visual occlusion, auditory ambiguity buildup, cross‑modal conflict, interoceptive perturbation (CO₂, somatic amplification), and individualized threshold‑mapping. This working paper advances integrative models of stress response, panic maintenance, freeze dynamics, and ambiguity‑driven threat inference, offering a computational and evolutionary lens for researchers in behavioral psychology, psychophysiology, cognitive science, computational psychiatry, VR threat simulation, and real‑time decision‑making under uncertainty.