Search whitepapers and bibliography records.

Although research on AI safety is advancing rapidly, there remains a critical shortage of interfaces specifically designed for conversational AI systems to address domestic violence. This paper seeks to highlight this gap. Domestic violence contexts often involve hidden surveillance risk, coercive monitoring by another person, fear of escalation, financial, legal, immigration, parenting, or housing dependency, emotional attachment to the abusive partner, gradual disclosure rather than explicit crisis declaration, danger created by visible support-seeking behavior, and risk that digital traces may be discovered. Therefore, a domestic violence safety framework cannot simply copy self-harm or mental-health crisis workflows. It requires a specialized interface logic that treats privacy, subtlety, autonomy, and proportional escalation as central safety requirements.