Role-Based Access Control (RBAC) is the standard model for enterprise authorization. However, automated role assignment typically prioritizes administrative efficiency over security. Traditional approaches minimize the number of roles, often producing coarse-grained policies that violate the Principle of Least Privilege by bundling high-risk permissions with common permissions. In this paper, we propose a risk-aware role allocation framework that explicitly accounts for permission sensitivity. We formulate the problem as a submodular cost minimization task and introduce a non-linear penalty for high-risk permissions. To address the resulting NP-hard problem, we develop a scalable greedy role allocation algorithm that exploits submodularity. We also consider budget-constrained scenarios, ensuring effective risk reduction under limited administrative capacity. Experimental results show that our approach scales almost linearly, achieves near-optimal solutions, and effectively isolates high-risk permissions where traditional methods fail.