Agentic AI for Child Welfare Intake and Early Risk Identification: A Governance-Aware Framework

Child welfare intake is the first point where someone raises a concern about a child, and the system receives the data and decides what to do next. Child welfare intake is where everything begins. The first few decisions taken here often shape the entire journey of a case. But in reality, this stage is messy, information comes in bits and pieces, decisions depend on individual judgment, and important details are sometimes missed. When this happens, vulnerable children may not receive the right attention at the right time. In this paper, we present a governance-aware conceptual framework and proof-of-concept implementation of a multi-agent artificial intelligence (AI) system designed to support child welfare intake and early risk identification. We have envisioned a system where the data entry will be done using a conversational interactive chatbot interface, and the system will have multiple agents to perform specific tasks such as analyzing the data, assessing risk, checking data quality, and monitoring bias. Also, there will be a Central Orchestrator, which will manage all the agents and maintain the sequence of the workflow. A couple of important ideas we've planted in our proposed system are that if the system is not confident about the output it generates, it does not propagate to the next phase; instead, it passes the case to a human in the loop. Also, the system asks human-friendly questions if the response from the user is incomplete. These features make the system more thoughtful rather than just rushed, and they help ensure that uncertain or high-risk situations are not handled only by automation. Our intention through this article is not to reduce human involvement in the child welfare system or to take over the role of caseworkers. Our main goal is to propose an agentic AI-enabled support system for caseworkers that has the potential to improve the consistency, completeness, and governance of child welfare intake processes. With appropriate real-world validation, the proposed system may play a supportive role in helping vulnerable children receive attention earlier.