arXiv cs.CR·4d agoRouxii: Exploiting Honeypots with Deception-Aware AI Pentesters#honeypot#llm-agents#pentestingAI safety & security