This research introduces new privacy paradigms for sequential decision-making AI systems where sensitive information emerges from patterns over time, not just isolated data points.

• Identifies unique privacy challenges in reinforcement learning (RL) applications, especially in federated RL and RLHF for large language models
• Explains how temporal patterns and behavioral strategies create novel privacy vulnerabilities
• Proposes frameworks that better protect privacy in collaborative AI learning environments
• Highlights critical implications for high-stakes domains like healthcare and finance

For security professionals, this work addresses fundamental gaps in current privacy protection approaches as AI systems become increasingly deployed in sensitive contexts.

Position Paper: Rethinking Privacy in RL for Sequential Decision-making in the Age of LLMs