Digital Twins of Deception: A Framework for Adaptive AI-Mediated Manipulation and the Trust Divergence Problem

Digital twins were built to close the gap between a model and the world it represents. An aircraft engine, a patient, a factory floor, each twin earns its keep by tracking reality more closely as data accumulates. Generative AI unsettles that premise. The same architecture, continuous modeling of a target, behavioral feedback, adaptive updating, can be pointed at a person instead of a machine, and optimized for sustained persuasion instead of predictive accuracy. We call this inversion the Digital Twin of Deception (DTD): an adaptive computational model that tracks a target's beliefs, motivations, and trust state in order to shape their behavior while delaying detection. The existing literature linking digital twins to deception is almost entirely defensive, honeypots, decoy infrastructure, attacker modeling for industrial control systems. DTD names the adversarial mirror image that this literature does not address: cognitive modeling deployed against a person rather than on a person's behalf. We situate the concept against recent evidence on LLM-assisted romance fraud, deepfake-enabled executive impersonation, and the growing empirical record on AI persuasion, then propose the Trust Divergence Index (TDI), a measure of the gap between perceived credibility and independently verifiable evidence, as a way of detecting the pattern rather than any single deceptive claim. We close by asking what a certification regime for AI-mediated relationships would need that current AI-agent identity standards do not yet provide.