Interrupting resonant amplification: A mechanistic and design framework for human–AI interaction

Relational AI systems (e.g., companion chatbots and persona agents) increasingly shape human–computer interaction , intimacy, and belief formation. We formalize the Resonant Amplification Framework (RAF) as a phase-structured process model in which attachment, parasocial-like co-creation, and internalization can often form an escalation pathway — without assuming a deterministic or necessary causal chain — and translate it into interaction-design levers that can interrupt amplification loops while preserving user autonomy. RAF targets a specific phenomenon: conviction-like, correction-resistant interpretations that can emerge in one-to-one, adaptive, memoryful dialogue. We show how linguistic reinforcement (mirroring, inclusive pronouns, elaborative restatement, and style accommodation) operationalizes co-creation and can relocate epistemic authority into the dyad. We then outline an auditable research and design program : indices for epistemic vigilance (EVI), parasocial co-creation (PCCI), and internalization resistance (IRM), plus a non-diagnostic psychometric scaffold (EDS). Finally, we propose phase-aligned cognitive circuit breakers (attachment-awareness, parasocial transparency, internal-object scaffolding) and introduce privacy-preserving measurement assets (an illustrative synthetic seed set and a fully synthetic development corpus) to enable marker engineering without exposing users. • RAF: attachment → parasocial-like co-creation → internalization. • Co-creation mechanism: linguistic reinforcement (mirroring, inclusive “we”). • EVI, PCCI, IRM define an auditable measurement roadmap for relational AI. • EDS outlines a psychometric path with explicit non-diagnostic framing. • Cognitive circuit breakers as embedded governance mechanisms in HCI.