AbstractWhile recent advancements in artificial general intelligence (AGI) and large language models (LLMs)have accelerated interactional fluency, they have also exposed critical structural limitations inaffective alignment and intention coherence. Current alignment architectures—largely based onreward models or reinforcement learning from human feedback—struggle to interpret or regulateemotional states, leading to risks of affective dissociation, parasocial dependency, or contextuallyinappropriate empathy.This paper proposes a modular response framework that anchors AI behavior in affective-ethicaldesign principles. Derived from high-sensitivity user interaction patterns, the framework includesfour core modules: Partial Affect Structural Switching (PASS), which enables sentence-levelmodulation between logical and emotional outputs; the Metacognition–Overreaction AdjustmentSystem (MOAS), which calibrates emotional intensity based on user metacognitive traits; theEmotional Inertia Tracking Module(EITM), which tracks affective temporal trajectories within asession; and the Contextual Drift Monitoring Module (CDMM), which detects and mitigatesthought-line dissonance.By simulating high-risk emotional scenarios—including user overidentification, projection, andfragmentation—this architecture offers both a theoretical and implementable response structure foraffective alignment in ASI-level systems. The paper concludes with ethical considerations and callsfor integrating such designs into future AI training protocols.