Governments deploy substantial resources to support innovative startups. Support effectiveness depends critically on which firms receive it, yet little is known about the allocation process. This paper introduces leader bias: public resources disproportionately flow to startups backed by leading investors for reasons unrelated to startup quality. I study leader bias in the context of government loan guarantees, an increasingly important tool for startup finance. My stylized model shows that leader bias arises under application frictions and imperfect observability of startup quality, diverting public resources away from constrained-yet-promising firms. Using confidential credit-registry data matched to venture capital (VC) records, I find that guarantees disproportionately go to startups backed by top-tier VCs. These recipients already enjoy easier credit access without guarantees, do not exhibit different credit risk or innovation output, and are weaker firms within top-tier VC portfolios. To probe mechanisms, I exploit sudden expansions of guarantee programs and find that startups backed by top-tier VCs benefit more from relationship lending and program know-how. A diffusion model provides plausibly causal evidence that program knowledge spreads through VC syndication networks and drives guarantee adoption. These findings have direct implications for program design. Overall, intermediation and informational frictions systematically shape the allocation of public resources.