Discussions of artificial intelligence (AI) and biological weapons (BW) have largely focused on laboratory-facing capabilities, treating acquisition as a technical problem centred on pathogen design, experimental acceleration, or optimization. This emphasis risks compressing the work of BW acquisition into a narrow set of moments, obscuring the wider labour through which harmful capability has historically been pursued, sustained, and often abandoned. Drawing on the historical record of BW programmes, this paper starts from the observation that acquisition has never been reducible to laboratory development alone. Instead, it has unfolded through extended processes of organisation, interpretation, coordination, and constraint, spanning multiple stages and shaped by uncertainty, failure, and adaptation. To capture this, the paper introduces the concept of the labour of harm to describe the situated work through which harmful intention is translated into operational capability. The article deploys the web of acquisition—a seven-stage analytical framework encompassing inspiration, planning, development, manufacture and production, storage and transportation, targeting, and dissemination—to conduct a sustained, stage-by-stage examination of the distinct forms of labour involved in BW acquisition, and of how different AI-enabled tools may intersect with, support, or complicate that labour at each point. This analysis moves beyond diagnosing gaps in existing discussion to examine how AI may shape imagination, coordination, experimentation, scaling, concealment, and execution in ways that vary markedly across stages. This stage-specific analysis shows that AI does not enter acquisition as a single, uniform influence. Instead, its potential significance lies in how it may unevenly redistribute, reconfigure, or intensify labour across the acquisition process, without dissolving the organisational, material, or environmental constraints that have historically shaped BW programmes. Understanding the implications of AI for BW therefore requires attention to process, contingency, and organisational endurance, rather than a narrow focus on laboratory feasibility alone.