Catastrophic risk mitigation and AI safety evaluation through threat modeling, scenario analysis, and adversarial testing.