Anticipating Drifts in Protocol Design through Adversarial LLM Simulation

Botao Amber Hu, Helena Rong

Keywords: Protocol Design; Synthetic Personas; Social Simulation; LLM Agents; Second-Order Dynamics; Design Futuring; Governance

Abstract: Governance protocols---the rules, norms, and technical standards that coordinate collective action---are notoriously difficult to evaluate before deployment. Once enacted, they may \emph{drift} from intent, become \emph{jammed} by adversarial actors, or \emph{ossify} against necessary change. These second-order dynamics emerge from the situated reasoning of actors who were not in the room when the rules were written, making them invisible to conventional testing. We argue that large language models, grounded in deep synthetic personas, can serve as a pre-deployment stress-testing mechanism for governance protocols---surfacing identity-specific failure modes that neither classical agent-based modeling nor participatory design workshops can reach alone. We operationalize this argument through \textbf{ProtoPoliSim}, a working prototype that instantiates persona-driven agents inside the \emph{Protocol Futuring} adversarial cycle---a structured Blue~(build) / Red~(attack) / Temporal~Shift / Observer pipeline---to automatically elicit and classify second-order dynamics as structured simulation outputs. Each run produces a branching tree of protocol futures, with every node annotated for drift, jam, ossification, or reinterpretation. We present the system's architecture, demonstrate its interface, and discuss the open methodological and validation challenges this approach poses for the PoliSim community.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 12