AI-enabled biology is increasingly a connected system of biological databases, computational pipelines, AI models and agents, human decision makers, and physical biological entities and processes. Yet many approaches to AI-biosecurity still evaluate individual components or capabilities rather than asking how security emerges from the relationships among them.

In this talk, I will introduce relational biosecurity and an emerging research program aimed at making these relationships objects of empirical study. Beginning with technical relationships within AI-bio workflows, I ask how information, context, provenance, control, and feedback move across interfaces; when errors and other perturbations propagate or are buffered; and which system properties enable detection, attribution, correction, and recovery. I will discuss approaches including relationship mapping, structured observation of existing AI-bio workflows, controlled evaluation of technical relationships, and the longer-term possibility of field surveillance infrastructure for understanding how AI-bio systems behave in practice. I will also consider how empirical evidence about these relationships could inform biosecurity policy, system design, and security training. The broader goal is to move beyond asking whether individual models or tools are safe, toward understanding how security and resilience emerge in connected AI-enabled biological systems.