Risk analysis is a prerequisite for the design and implementation of risk-reduction defenses. In the area of dynamic autonomously-adapting systems, especially those that interact physically with human beings (e.g., service robots), current risk analysis techniques (fault trees, failure modes and effects analysis, hazard operability (HAZOP), etc.) are quite limited and difficult to relate to the development process. We therefore investigate which system behavior models and human-system interaction models are appropriate for early risk analysis of autonomously-adapting systems. We particularly focus on modeling languages like UML associated with risk analysis methods. Our research also focus on the development of tools for supporting such analysis.