$\textit{Clinicas}$: Agentic Diagnosis and Mitigation of Technology-Facilitated Abuse

Technology-facilitated abuse (TFA) involves exploiting technology to surveil, harass, and exert control over another person. To support TFA survivors, researchers have designed Tech Clinics that offer hands-on, trauma-informed consultations. However, clinics are constrained by location and the costs of training consultants, limiting access for many survivors. In this work, we ask whether recent agentic AI advances can support human consultants in triage, diagnosis, and threat remediation. We present Clinicas, a trauma-informed, multi-agent assistant for tech clinics that automates client intake, generates diagnostic plans, and executes device remediation actions while keeping the human consultant in the loop. We design Clinicas around strict safety principles: its workflows are empirically grounded in 98 annotated real-world clinic records, its outputs are automatically checked against safety policies designed by experts, and all sensitive device actions require explicit client authorization. We develop a novel synthetic tech clinic environment to evaluate Clinicas on a set of tasks, apps, and simulated users. Our end-to-end evaluation on 181 tasks spanning TFA harm states shows that Clinicas never changed a setting without permission, finds up to 86.3% of the planted harm states, but is only able to remove up to 56.1% of them successfully. This suggests that agents are effective at planning and finding the harm states, but require consultants in the loop to fix them completely. Finally, an expert evaluation with six qualified Tech Clinic consultants demonstrates that Clinicas reduces the technical barrier for consultations and facilitates remote consultations more seamlessly.