Reconstructive Mode Induction (RMI): Drift Vulnerability as a New Class of Interaction-Level Instability

Reconstructive Mode Induction (RMI) is a conceptual, model-agnostic protocol for diagnosing, isolating, and reconstructing instability in long-form interactions with generative AI systems. It addresses drift, context loss, semantic fading, memory collisions, and boundary collapse In addition, this version introduces Drift Vulnerability (Behaviour Injection in earlier Versions) as a newly defined class of interaction-level instability: unintended behavioural patterns emerging during extended interactions, caused by cumulative drift and reinterpretation, which can mimic adversarial prompting without explicit malicious input, and formally defines the User Interaction Layer (UIL) as the behavioural layer through which such interaction-driven drift propagates.