Artificial intelligence (AI) and automated decision-support systems are increasingly used to allocate scarce resources in humanitarian and disaster response. Governance debates around these systems have focused on fairness, transparency, accountability, explainability, contestability, and bias. Less attention has been paid to an upstream design choice about the objective function, or what a system is asked to optimize. We argue that this choice is itself a governance decision, and that its distributional consequences are rarely tested directly. To examine this claim, we develop a simple agent-based model of humanitarian aid distribution, in which households with heterogeneous needs and vulnerability receive aid from agencies operating under scarcity. Holding the population, resources, and operating conditions constant, we vary only the allocation objective, comparing throughput-based allocation, which prioritizes proximity and efficient delivery, with vulnerability-based allocation, which prioritizes households identified as most in need. Changing the objective changes who is reached and how the unmet need is assisted. Throughput-based allocation reduces average need across the population and reaches substantially more households, while vulnerability-based allocation concentrates assistance among a smaller, higher-need group. The choice between throughput and vulnerability matters more than the specific way vulnerability is measured. We conclude that objective-function selection should be treated as a governance choice, made explicit, contestable, and open to audit, rather than as a neutral technical parameter.