The Agentic AI Army That Never Was: Projecting LLM Swarm Narratives with ‘Noisy’ LLM Sock Puppets and Whaley’s Theory of Outs

Abstract

This short position paper suggests there may be greater deception and influence of an attacker’s perceptions of a fictional ‘Agentic AI Army’ swarm of LLM sock puppet network defenders than deploying real LLM agent swarms. We model a counterintuitive industry approach integrating Whaley’s lesser-known Theory of Outs and “turnabout” deception techniques to encourage a human or LLM attacker’s discovery of deception on an industry network. While we recognize that the knowledge of real or imagined deception can deter an attacker, we also recognize that attackers may demonstrate greater confidence on a network after discovering what appears to be deception artifacts. We visualize how ‘noisy’ LLM sock puppets inside of a network that prompt optimized query returns on their content and placement on the network could draw attackers to later stage deception functions and effects and enhanced defender alerting and analysis on human or LLM attacker interaction with those deception functions. We find in anecdotal operational research that highlighting ‘noisy’ sock puppet content enhances high-fidelity detection. We frame these findings using this integrated industry model in the context of LLM swarm narratives for deception. There has been an increasing concentration on swarming as a military technique and military strategy, as modern military conflicts continue to adapt to irregular warfare environments. The renewed concentration on developing and integrating swarm intelligence with LLM agents continues to face limitations, in terms of simulating natural swarm behaviors and operating autonomously as part of a decentralized model. This short position paper proposes a more immediate deception and influence effect, namely projecting fictional LLM swarm narratives suggesting there is an ‘Agentic AI Army’ assisting human defenders. We use organizational perception management as a design framework to visualize a deception and influence narrative communicating this fictional narrative using ‘noisy’ LLM sock puppets and our integrated model of Whaley’s Theory of Outs and “turnabout” deception techniques.