Agentic AI for Offensive Security: LLM-guided Autonomous Red Teaming in a Limited Cyber-range Environment
Results show that, within a controlled cyber-range environment, LLM-guided agents can approximate deterministic performance when appropriate constraints are applied, suggesting that agent-based approaches to offensive security can support semi-autonomous red teaming workflows, provided that decision-making is governed by structured control policies.