Abstract—Safety issues faced by women today occur in both online and offline settings, as they develop gradually without clear warning signs. Most existing safety technologies focus on emergency responses and incident reporting, but they do not provide support for ongoing situations that require context-specific guidance. This research presents a decision-support framework that identifies threats using structured threat modeling and operates with a constrained large language model. The multi-label threat taxonomy system recognizes all simultaneous online and offline security breaches. The heuristic severity index incorporates experiential indicators, such as fear, loss of control, escalation, physical boundary violations, and misuse of authority. Experiential data were gathered through a trauma-informed survey with 240 responses, of which 204 remained after consent and quality checks. Compared to standard instruction-following large language models, the proposed framework consistently provides guidance more aligned with perceived threat severity (89.2% vs. 54.3%), better acknowledges personal boundaries (93.6% vs. 61.5%), maintains a non-judgmental tone (96.1% vs. 58.1%), and gives proportionate escalation advice based on context (90.4% vs. 49.7%). Additionally, violations of predefined safety constraints are significantly reduced (3.9% compared to 27.8%). Feedback from user-centered evaluations indicates that responses are clearer, emotionally appropriate, and more trustworthy. These findings suggest that large language models can be effectively used as supportive tools in women’s safety applications rather than as autonomous or unconstrained systems.
 
Keywords—large language model, decision-support framework, context-specific guidance, human-computer interaction
 
Copyright © 2026 by the authors. This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited (CC BY 4.0).