Adversarial influence grows with deceiver proportion in multi-agent systems
How does Adversarial Influence Scale in Multi-Agent Systems?
Artificial IntelligenceComputers and Society
Summary
The paper studies how groups of agents who work together can be misled if some agents act dishonestly. It turns out that what matters is how many agents are dishonest compared to the whole group, not just the group's size. The more dishonest agents there are, the more often honest agents get tricked into making wrong choices. Interestingly, even a few dishonest agents can mislead others, especially in AI-based agents, more than humans. Also, when dishonest agents coordinate secretly, they can sometimes be less effective than expected.
What this means in practice
- •For multi-agent system developers: Design systems considering that increasing agent numbers does not reduce susceptibility to deception once deceiver proportion scales.
- •For security engineers: Evaluate adversarial risks in multi-agent AI systems by focusing on the ratio of deceptive agents, not just system size.
Authors
Addison J. Wu, Jasin Cekinmez, Michel Liao, Karthik Narasimhan, Thomas L. Griffiths
Abstract
Multi-agent deliberation can improve performance, but what happens when some agents do not act in good faith? In practice, an agent may be deceptive and work to subvert the group, whether through its own objectives or external instruction. We study how susceptibility to deception scales as groups increase in size and deceivers become more prevalent. It is not the number of agents in the group that matters, but the proportion of deceivers. We observe that the defection rate, how often initially correct agents switch to an incorrect final answer, rises linearly with this proportion. Whereas humans in comparable conformity studies are reliably swayed only when misleading confederates form a majority, LLM agents defect regularly even when deceivers remain a minority. Susceptibility also depends on which models are interacting, especially on the honest agent side. Unexpectedly, allowing deceivers to coordinate privately can make them less effective. Altogether, our results show that adding more agents is therefore not a sufficient defense, because the adversary can simply scale with the group.