PrivEscalate Scales Linux Privilege-Escalation Evaluation From Under 15 Scenarios to 531 Dockerized Ones

arXiv 2609.09087·medium signalXBlueskyLinkedInCopy link

Prior LLM privilege-escalation evaluations used fewer than 15 scenarios; PrivEscalate provides 531 Dockerized scenarios across 14 sub-categories plus 329 parameterized variants that measure sensitivity to environmental distractors. Evaluating six LLMs across three agent architectures shows capability is heterogeneous across vulnerability classes with no single model dominating the high-prevalence ones, that successes are sensitive to environmental perturbation so configuration rotation disrupts some attempts without eliminating risk, and that the agent architecture itself materially changes success rates and reorders model rankings. The authors also release PrivEscAgent, a wrapper adding deterministic enumeration, category matching and step planning to a generic ReAct agent.