Can ethical artificial intelligence aid infant and young child feeding support during conflict? A review of the experience and lessons learned from FHI 360′s in Ukraine

Background: , a closed-domain, ethically governed Artificial Intelligence (AI)-enabled breastfeeding and Infant and Young Child Feeding in Emergencies (IYCF-E) support system, integrated with facility-based counseling and a national mentorship programme for lactation counselors. Objectives: The aim of this study was to evaluate the feasibility, safety, acceptability, and operational contribution of an AI-enabled hybrid counseling model to continuity and quality of Infant and Young Child Feeding in Emergencies services during protracted conflict, including its integration with facility-based counseling and mentorship systems and its potential influence on caregiver feeding practices and counseling quality. Methods: A convergent mixed-methods approach triangulated: (1) chatbot analytics (user characteristics, interaction volume, response quality); (2) facility counseling data from four conflict-affected oblasts; (3) structured competency assessments following >600 mentorship visits; and (4) qualitative feedback from caregivers and health workers. Governance, safety, and Code-compliance safeguards were assessed against the World Health Organization (WHO) AI ethics guidance, the Operational Guidance on Infant and Young Child Feeding in Emergencies (OG-IFE), and Ukraine's breastfeeding policies. Results: Between November 2024 and October 2025, 2,066 caregivers generated more than 38,000 chatbot interactions. Ninety-eight percent of answers met accuracy and safety criteria; one hallucination event was detected and corrected through real-time review. Caregivers used the system predominantly during periods of insecurity or when services were inaccessible, valuing its 24-h availability, emotional reassurance, and escalation to human counselors. In parallel, facility data showed improved quality of counseling, strengthened referral pathways, and increases in early initiation and exclusive breastfeeding among women receiving repeated support. Mentorship visits demonstrated competency gains among lactation counselors, enhanced adherence to the Code, and more consistent use of MoH-aligned counseling tools. Conclusion: An AI-enabled, human-supervised hybrid model is feasible, acceptable, and safe for sustaining breastfeeding and IYCF-E support during active conflict. When anchored in authoritative guidance and embedded within national systems, AI tools can complement skilled counselors, strengthen continuity of care, and uphold Code-compliant, evidence-based support for mothers and infants. To our knowledge, this is the first documented evaluation of an AI-enabled IYCF-E intervention implemented during an active conflict.