Full Description
Online grooming poses a serious threat to youth safety, often unfolding through gradual conversational stages that manipulate trust and personal boundaries. While prior work has primarily focused on detection and moderation, proactive education remains underexplored.
We present ALLY (Adaptive LLM for Learning and Youth-safety), an educational AI system specifically designed to teach safe and context-appropriate responses in cybergrooming scenarios. ALLY leverages stage information and conversational context to generate adaptive feedback and micro-tutoring that foster awareness, boundary-setting, and resilience. Built on a stage-guided tutoring framework, ALLY combines LoRA-based fine-tuning on tutoring-style response data with retrieval-augmented knowledge grounding from protective communication strategies, enabling psychologically supportive and stage-consistent feedback.