AI negotiation agents are rapidly proliferating on online spaces, brokering millions of transactions monthly between businesses and consumers. As platforms increasingly endow these agents with distinct personality traits, understanding how personality shapes fairness, exploitation dynamics, and linguistic toxicity becomes critical for platform governance and consumer protection. As many are endowed with distinct personality traits, understanding how these traits shape negotiation behavior is crucial. Using canonical OCEAN personality definitions, we initialize LLM agents and simulate single- and multi-issue bargaining games (n = 3,500) within a game-theoretic framework. Results reveal two key insights. First, we identify the linkage between personality, fairness, and exploitation: agreeable agents have the highest payoffs but are exploitable by less agreeable agents. Second, agents tend toward fairness rather than rationality, which may arise from innate guardrails. However, guardrails can be "jailbroken" with the right personalities. Personality thus shapes both quantitative (surplus discovery) and qualitative (toxicity, persuasion) dimensions of negotiation. We discuss how AI personality steers negotiation behavior toward certain domain objectives, with practical implications for designers.

PDF