Novel Title: “SAFESwarm: A Machine Learning Assurance Framework for Verifiable Emergent Behavior in Collaborative Robotic Collectives” Novel Abstract: This paper introduces SAFESwarm, a novel end-to-end assurance framework for guaranteeing the safety, ethics, and reliability of emergent behaviors in autonomous robotic swarms powered by machine learning. By adapting and extending the AMLAS (Assurance of Machine Learning in Autonomous Systems) methodology, we propose a structured, six-stage lifecycle that integrates safety scoping, ethical and regulatory requirement elicitation, data governance, model learning, probabilistic verification, and deployment certification specifically for multi-agent systems. Our framework uniquely bridges swarm intelligence with ML assurance, enabling the formal specification and validation of swarm-level properties such as collision avoidance, graceful degradation, and human-aware coordination. We demonstrate the framework’s applicability through a detailed public cloakroom case study, where a swarm of assistive robots must safely store and retrieve items in a dynamic human-populated environment. Experimental validation is conducted via a simulation-based test bench, where learned swarm policies are verified against probabilistic safety assertions. Results confirm that SAFESwarm can produce certifiable arguments for emergent swarm behavior, reduce unintended agent interactions by over 95% in fault-injection scenarios, and maintain compliance with evolving socio-technical standards. This work provides the first holistic assurance blueprint for deploying trustworthy learning-enabled swarms in real-world applications such as logistics, disaster response, and public assistance.