As a key roadside safety feature, longitudinal guardrail steel barriers are purposefully designed to contain and redirect errant vehicles to prevent roadway departure, dissipate impact energy through plastic deformation, and reduce the severity of vehicle crashes. Nevertheless, these systems should be carefully designed and assessed, as localized rupturing, especially near splice or impact locations, can lead to catastrophic failures, compromising vehicle containment, violating crash safety standards, and ultimately jeopardizing the safety of occupants and other road users. Before conducting full-scale crash testing, finite element analysis (FEA) tools are widely employed to evaluate the design efficiency, optimize system configurations, and preemptively identify potential failure modes prior to expensive physical crash testing. To accurately assess system behavior, calibrated material models and precise failure criteria must be utilized in these simulations. Despite the existence of numerous failure criteria and material models, the material characteristics of AASHTO M-180 guardrail steel have not been fully investigated. This paper significantly advances the FE modeling of ductile fracture in guardrail steel, addressing a critical need within the roadside safety community. This study formulates stress-state-dependent failure criteria and proposes advanced material modeling techniques. Extensive experimental testing was conducted on steel specimens having various triaxiality and Lode parameter values to reproduce a wide spectrum of complex, three-dimensional stress-state loading conditions. The test results were then used to identify material properties and construct a failure surface. Subsequent FEA, which incorporated the Generalized Incremental Stress-State-Dependent Damage Model (GISSMO) in conjunction with two LS-DYNA material models, illustrates the capability of the developed surface and material input parameters to predict material behavior under various stress states accurately. A parametric study was completed to further validate the proposed models, highlighting their robustness and reliability.