Panel
In person
LeadersGoldDiscovery

Defining and scaling beneficial model behaviors assessment

  • Date
    9 July 2026
    Timeframe
    14:30 - 15:00
    Duration
    30 minutes
    • Days
      Hours
      Min
      Sec

    As conversational AI systems become increasingly relational, understanding their impact on childrens cognitive, emotional, and social development has become a critical challenge for developers, educators, researchers, and policymakers.

    During this session, everyone.AI will present AïA Safety Builder, their AI-powered behavioral assessment technology designed to evaluate relational AI systems at scale. The platform analyzes interactions between AI models and young users, identifying anthropomorphic, intersectional and relational behaviors associated with socio-affective pull that increase risk of emotional over-reliance and unhealthy attachment to AI systems across educational, entertainment, and emotional support contexts. At the core of the approach is CALIBER, a developmental framework that structures expert consensus around healthy and potentially problematic relational behaviors in human-AI interactions. This framework is operationalized through SRL4Children, an AI-powered assessment engine that combines expert-defined criteria with calibrated LLM-based evaluation to perform large-scale behavioral analysis and generate contextualized risk assessments.

    The session will demonstrate how AI itself can help transform qualitative concerns about relational AI into measurable and actionable indicators. The long-term objective is to contribute to the emergence of an independent quality and trust framework for childrens AI systems, similar in spirit to safety and quality labels that help families, educators, and institutions make informed choices about products intended for young audiences.

    Share this session with your network
    • Session starts in
      Days
      Hours
      Min
      Sec

    Are you sure you want to remove this speaker?