Local-inference language models structurally lift the supervisory bottleneck that has historically constrained influence operations. We formalize SAINTS (Synthetic Autonomous Identity Network for Targeting with Saturation), a class of infrastructure in which each persona is carried by a physically isolated node running a model fine-tuned on the persona's behavioral profile at the weight level. SAINTS is introduced here as a formal threat model rather than an implemented or experimentally validated system : the contribution is not a deployable specification, it is a structural threat model precise enough that its breaking surfaces become analyzable. We characterize four architectural properties (non-correlation, embodied behavioral coherence, opaque action chains, saturation), formalize the cost asymmetry between deployment and investigation, and identify five detection surfaces that the architecture cannot absorb. The central analytical claim is a structural inequality : under the modelling assumptions stated, investigation cost grows super-linearly in network size while deployment cost grows linearly. We characterize a threshold parameter 𝑘* beyond which investigation becomes, in this model, economically irrational ; its precise numerical value depends on the adversary-defender configuration and the numerical estimates given here are illustrative, not normative, and require empirical calibration. Patience, defined as waiting for unavoidable human-operator failure, appears in this analysis as one of the principal defensive vectors whose cost scales symmetrically with attacker deployment.
Charles Mordelet (Thu,) studied this question.