EvoEmo: Towards Evolved Emotional Policies for Adversarial LLM Agents in Multi-Turn Price Negotiation

Researchers present EvoEmo, an evolutionary reinforcement learning framework that enables LLM agents to develop dynamic emotional strategies in multi-turn price negotiations. The system outperforms baseline approaches by achieving higher success rates and efficiency while improving buyer outcomes, demonstrating that adaptive emotional expression enhances AI negotiation capabilities.

EvoEmo addresses a fundamental gap in LLM agent development: the inability to strategically deploy emotions during adversarial negotiations. While recent advances in Chain-of-Thought reasoning have enabled complex multi-turn interactions, existing agents remain passive in emotional expression, making them susceptible to manipulation. This research bridges that gap through evolutionary reinforcement learning that models emotional state transitions as a Markov Decision Process, allowing agents to learn when and how to express emotions for maximum negotiation advantage.

The framework's innovation lies in its population-based genetic optimization approach, which evolves emotion policies across diverse scenarios rather than applying static emotional responses. By treating emotions functionally—as negotiation tools rather than mere reactions—the research reconceptualizes how AI agents can operate in strategic interactions. This aligns with broader trends in agentic AI that prioritize adaptability and context-awareness over rigid rule-following.

For the AI development community, EvoEmo demonstrates measurable improvements: higher success rates, improved efficiency, and increased savings for buyers compared to vanilla and fixed-emotion baselines. These results validate that emotions serve computational purposes in negotiations, not just social functions. This has implications beyond price negotiation—strategic emotional deployment could enhance AI performance in contract discussions, conflict resolution, and complex business interactions.

Future development should examine whether evolved emotional policies transfer across different negotiation contexts and adversary types. The open-source code release enables community validation and extension, though practitioners must consider whether LLM-generated emotional expressions in real negotiations raise ethical or transparency concerns.

• →EvoEmo uses evolutionary reinforcement learning to optimize dynamic emotional strategies in LLM-based price negotiations.
• →The framework outperforms static emotion and vanilla baseline approaches across success rates, efficiency, and buyer savings metrics.
• →Emotional expression functions as a strategic negotiation tool rather than a passive response, improving agent resilience against adversarial counterparts.
• →The research demonstrates that emotions serve computational purposes in multi-turn negotiations, extending AI capability beyond traditional reasoning frameworks.
• →Open-source code availability enables reproducibility and community-driven advancement in emotion-aware LLM agent design.