PAPER·May 29, 2026How Reliable Are AI Attackers Against a Fixed Vulnerable Target? A 400-Run Empirical Study of LLM Penetration Testing ConsistencyarXiv
