How does Adversarial Influence Scale in Multi-Agent Systems?

Source facts

The preprint studies deception in multi-agent deliberation. Its abstract reports that susceptibility depends on the proportion of deceivers rather than total group size, that defection rises approximately linearly with deceiver proportion, and that model pairing matters. It also reports that private coordination can reduce deceiver effectiveness, while adding agents alone is not a sufficient defense.

Analysis

The result frames multi-agent robustness partly as a composition and communication problem: group size alone may not determine resilience when adversarial agents are present.

Uncertainty/Doubt

• This is an arXiv preprint. The result is specific to the tested deliberation setup and should not be treated as a prevalence estimate for deployed multi-agent systems.