This paper proposes a graph-based method for forming corporate-network security domains from object-level risk attributes. Importance, vulnerability, attack probability, and estimated loss are combined into a continuous risk score and mapped to discrete security levels. Objects with equal levels are grouped into disjoint domains, mixed-level service conflicts are resolved, and boundary controls are inserted on inter-domain edges. A Python prototype implements the scoring, graph partitioning, and boundary-insertion sequence. The method is evaluated on a synthetic 150-node corporate topology with 5,000 daily flows. Relative to a flat baseline, the reported scenario reduces average blast radius from 100% to 18.5% and directed attack paths from 22,350 to 1,240. Unlike VLANs, which provide logical network separation, Zero Trust, which applies per-request policy decisions, and SDN, which provides programmable forwarding control, the proposed method computes risk-based domain membership and boundary locations. It is therefore complementary to these enforcement approaches. The study is a simulation and prototype evaluation rather than a production-network deployment.