arXiv:2607.05398v1 Announce Type: new Abstract: Personas are often employed to guide large language model agents, yet their effectiveness in shaping strategic behavior in social dilemma settings remains uncertain. To address this, we examined the impact of persona prompts in an iterated Split or Steal game where persona-driven agents interacted with a Virtual Human (VH) controlled by a fixed prompt. Agents were instantiated from four open models (Ministral 3:3b, phi4:14b, Gemma3:12b, and Gemma4:e4b) at two temperature settings (0.3 and 0.7) and deterministic decision with zero temperature, whi
The proliferation of sophisticated large language models necessitates a deeper understanding of how their behavior can be predictably influenced in complex social situations, especially through accessible methods like persona prompting.
Understanding how personas shape AI agent behavior in dilemma games is crucial for developing robust, ethical, and strategically aligned AI agents that can navigate real-world interactions effectively.
This research provides concrete evidence of how simple persona engineering can significantly alter AI agent decision-making in social dilemmas, offering a foundational toolkit for agent designers.
- · AI Agent Developers
- · Ethical AI Researchers
- · Organizations deploying AI for negotiation/interaction
- · Developers unable to control agent behavior
- · Unregulated AI agent deployments
Persona engineering becomes a critical skill for AI agent development, moving beyond simple instruction following to shaping strategic and social behavior.
The ability to predictably influence AI agent behavior through personas could lead to more nuanced human-AI and AI-AI interactions in complex operational environments.
This could enable the creation of highly specialized AI agents with tailored social 'personalities' for specific tasks, potentially blurring the lines between programmed and emergent strategic behavior.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.CL