socialprotection.org (12.12.2025) This paper considers the methodological challenge of mapping systemic risks and harms emerging from the use of algorithms and artificial intelligence in social welfare systems. Recent tragedies in social welfare put focus on the role of algorithmic systems in the execution of social security policies, motivating new governance and regulatory measures. While many efforts have tried to address risks at the level of the technology, individual process or organization, algorithmic risks in social welfare are inherently sociotechnical. Addressing these risks involves various actors, bringing in additional normative and political complexity. In this study, we apply two methods known for their ability to address parts of the complexity. Actor analysis is used to analyse the multi-actor aspect and associated normative dimensions, and a system safety analysis is used to map and analyze the sociotechnical nature and mechanisms of algorithmic risks. We motivate why and how these methods are combined and reflect on their synergy and challenges. The study is situated in the establishment of a Dutch algorithm watchdog, and focuses on the case of Dutch social security. As such, this study is a first of its kind to apply system safety to algorithms in the social welfare domain, and provides methodological contributions by using actor analysis to better scope and inform the multi-actor and cross-organizational nature of the safety analysis.