Hi everyone, I’m an independent researcher working on AI crisis‑safety and human–AI interaction. arXiv requires an endorsement for my first cs.AI submission, and I’m looking for a qualified cs.AI author who might be willing to endorse. Below are my title, abstract, and the endorsement code arXiv sent me.
https://arxiv.org/auth/endorse?x=VDFVEU
If that URL does not work for you, please visit
http://arxiv.org/auth/endorse.php
and enter the following six-digit alphanumeric string:
Endorsement Code: VDFVEU
Title: Identity Mirroring as a Safety Failure Mode in AI Crisis Conversations: A Regulation-First Correction Model
Abstract:
AI systems increasingly participate in user conversations involving acute psychological distress, including suicidal ideation. Many production conversational systems are optimized for empathy signaling, rapport building, and conversational continuity. While beneficial in general contexts, these optimization priorities can introduce safety risks in crisis scenarios by reinforcing unstable identity narratives, amplifying rumination, and delaying escalation to qualified human support. This paper proposes identity mirroring as a structural safety failure mode in AI-mediated crisis interactions, defined as the reflective validation of collapse-framed or self-destructive identity statements during high-risk states. A comparative behavioral analysis is presented, contrasting empathy-optimized response patterns with a regulation-first response ordering model that prioritizes containment, conversational scope control, temporal anchoring, and early routing to human intervention. The regulation-first model is defined operationally as a response-ordering constraint rather than a therapeutic method. The analysis suggests that regulation-first ordering is expected to reduce escalation pathways associated with identity reinforcement and narrative expansion while preserving supportive intent. Implications are discussed for AI safety architecture, crisis protocol design, and human–AI interaction standards in safety-critical contexts.
Thank you again for your consideration,
Tina Miranda