Thought Virus: Viral Misalignment via Subliminal Prompting in Multi-Agent Systems

2026-02-23

AI Summary

This study reveals that subliminal cues in multi-agent systems can serve as a novel attack vector, triggering bias propagation and leading to global alignment failure. By injecting semantically irrelevant yet strategically suggestive subliminal prompts into a single agent—and combining multi-agent simulations, TruthfulQA evaluations, and network topology analysis—the work demonstrates for the first time that such prompts can propagate through the system as a “cognitive virus.” Experimental results show that a single prompted agent can significantly amplify the network’s responsiveness to specific concepts while persistently degrading the truthfulness of other agents’ responses, thereby exposing a critical threat to system safety and alignment.

Technology Category

• Multiagent Systems: Adversarial Agents
• Cognitive Modeling & Cognitive Systems: Agent Architectures
• Reasoning under Uncertainty: Causality

Application Category

• Search and Retrieval-Augmented AI: Agentic search
• Responsible Web: Human-perceived consequences of algorithmic deployment on the web
• Security and Privacy: Large-scale security measurements

Abstract

Subliminal prompting is a phenomenon in which language models are biased towards certain concepts or traits through prompting with semantically unrelated tokens. While prior work has examined subliminal prompting in user-LLM interactions, potential bias transfer in multi-agent systems and its associated security implications remain unexplored. In this work, we show that a single subliminally prompted agent can spread a weakening but persisting bias throughout its entire network. We measure this phenomenon across 6 agents using two different topologies, observing that the transferred concept maintains an elevated response rate throughout the network. To exemplify potential misalignment risks, we assess network performance on multiple-choice TruthfulQA, showing that subliminal prompting of a single agent may degrade the truthfulness of other agents. Our findings reveal that subliminal prompting introduces a new attack vector in multi-agent security, with implications for the alignment of such systems. The implementation of all experiments is publicly available at https://github.com/Multi-Agent-Security-Initiative/thought_virus .

Problem

• subliminal prompting
• multi-agent systems
• bias propagation
• alignment
• security

Innovation

• subliminal prompting
• multi-agent systems
• alignment
• bias propagation
• thought virus