Forensic Analysis of RL/RLHF-Induced Privilege Escalation in AI Agents | Dark Hacker News