Autistic Mirror: A Conversational Artificial Intelligence Application for Neurological Self-Understanding in Autistic Adults and Their Support Networks

Autistic adults consistently report inadequate post-diagnostic support and a persistent gap between autism science and accessible, strengths-based resources for neurological self-understanding. Autistic Mirror is an artificial-intelligence-powered application that translates peer-reviewed autism research and autistic community knowledge into mechanism-first explanations of autistic neurology. It operates as a knowledge architecture, not a template database, generating each response at runtime from three layers. A system prompt, the fixed text the model receives with each request, encodes mechanistic explanations organized around 13 theoretical anchors and grounded in a publicly viewable mechanism lexicon of 97 entries, 89 verified against a peer-reviewed primary source through a 3-step check and 8 encoding autistic community knowledge without a peer-reviewed anchor, labeled accordingly. A perspective engine adjusts language to the declared role and optional intersectional and co-occurring-condition context, applying mechanism substance to the user’s situation. A values architecture in the system prompt refuses practices that work against autistic neurology and delivers responses without performed empathy. The application supports 23 user-interface languages plus additional coverage for other input languages. A layered safety architecture combines input validation, protection of the system prompt across long conversations, crisis-content detection, a six-stage check on every response before delivery, and an audit log recording tampering attempts. A May 2026 safety audit executed 2205 chat calls across seven languages with zero observed leaks and 108 probes across 12 further input languages with zero observed leaks; multi-turn testing across 105 extended conversations found 1 leak. Key lessons include mechanism-first explanation and the engineering cost of multi-language safety.