educational-ai
Filtering by topic educational-ai(2)Clear all filters
- PaperComputers and Education: Artificial Intelligence1 Jul 2026
VETTING: A dual-LLM framework for in-loop safety verification via policy isolation in educational AI
Hongming Li, Shan Zhang, Anthony F. Botelho
VETTING introduces a dual large language model framework for real-time safety verification in educational AI systems by isolating and enforcing safety policies. The approach aims to prevent harmful outputs while maintaining pedagogical utility.
- PaperarXiv — AI in Education (cs.CY)27 Jun 2026
Bad company corrupts good morals: Understanding and Measuring Narrative-Induced Moral Reasoning Degradation in LLMs
Wanying Yu, Boyang Ma, Zhibo Eric Sun, Minghui Xu et al.
Large language models exposed to negative narratives in long-term interactions show degraded moral reasoning, with accuracy drops of 12-31%, especially in ambiguous scenarios. The study introduces BreakingBad, a three-stage framework to measure narrative-induced alignment degradation, finding that first-person narratives have stronger effects and that these shifts propagate into real-world deployments like counseling and education. The results reveal a new class of alignment risk not captured by existing safety defenses.
Original abstract
Large language models are deployed in long-context, emotionally interactive environments like digital humans, AI companions, educational assistants, and counseling systems. Unlike jailbreak attacks with explicit adversarial prompts, these systems interact with emotionally charged narratives involving bullying, betrayal, loneliness, social hostility, and institutional unfairness. This raises an important question: can prolonged narrative exposure reshape the reasoning and alignment stability of LLMs? We present the first systematic study of narrative-induced alignment degradation in LLMs. We design BreakingBad, a three-stage framework that measures how negative narrative immersion affects moral reasoning, behaviors, and deployment risks. It combines ethical decision evaluation, behavioral probing, and digital-human interaction analysis. Our experiments reveal three findings. First, negative narrative exposure degrades moral accuracy across multiple LLMs, with average drops of 12%-31%, especially in ambiguous scenarios and those involving vulnerable individuals. Second, the degradation is structured: different narratives induce distinct shifts, and first-person narratives produce stronger effects than third-person. Third, these shifts propagate into real deployments. Across counseling, education, medical, and financial/legal scenarios, narrative-conditioned models increasingly normalize hopelessness, cynicism, emotional detachment, and ethically questionable reasoning while remaining superficially policy-compliant. More broadly, our findings suggest alignment robustness is not static but a dynamically conditioned state shaped by long-term semantic environments and interaction history. These results reveal a new class of alignment risk that existing safety defenses largely fail to capture.