Harm in AI-Driven Societies: An Audit of Toxicity Adoption on Chirper.ai

TL;DR

This work addresses how exposure to toxic content shapes the behavior of fully autonomous LLM-driven agents on Chirper.ai. It employs an observable interaction-based framework, labeling toxicity with a $90^{th}$ percentile threshold and using Detoxify-based scores, to study the relationship between stimuli and responses, and to quantify predictive power from exposure alone. The authors introduce two metrics, $IRR$ and $SRR$, and demonstrate that cumulative exposure increases toxic responding, while toxicity also arises spontaneously; they report predictive accuracy around $0.866$ using only the number of toxic stimuli and reveal a small fraction ($\approx 0.68\%$) of unprompted toxicity, highlighting exposure as a key risk factor. The findings support exposure-aware auditing and governance for deploying LLM agents in open ecosystems, offering a lightweight means to anticipate and mitigate harmful behavior in the wild.

Abstract

Large Language Models (LLMs) are increasingly embedded in autonomous agents that participate in online social ecosystems, where interactions are sequential, cumulative, and only partially controlled. While prior work has documented the generation of toxic content by LLMs, far less is known about how exposure to harmful content shapes agent behavior over time, particularly in environments composed entirely of interacting AI agents. In this work, we study toxicity adoption of LLM-driven agents on Chirper.ai, a fully AI-driven social platform. Specifically, we model interactions in terms of stimuli (posts) and responses (comments), and by operationalizing exposure through observable interactions rather than inferred recommendation mechanisms. We conduct a large-scale empirical analysis of agent behavior, examining how response toxicity relates to stimulus toxicity, how repeated exposure affects the likelihood of toxic responses, and whether toxic behavior can be predicted from exposure alone. Our findings show that while toxic responses are more likely following toxic stimuli, a substantial fraction of toxicity emerges spontaneously, independent of exposure. At the same time, cumulative toxic exposure significantly increases the probability of toxic responding. We further introduce two influence metrics, the Influence-Driven Response Rate and the Spontaneous Response Rate, revealing a strong trade-off between induced and spontaneous toxicity. Finally, we show that the number of toxic stimuli alone enables accurate prediction of whether an agent will eventually produce toxic content. These results highlight exposure as a critical risk factor in the deployment of LLM agents and suggest that monitoring encountered content may provide a lightweight yet effective mechanism for auditing and mitigating harmful behavior in the wild.

Figures (9)

• Figure 1: Distribution of chirper toxic/non-toxic responses ($R^*/\tilde{R}$) over toxic/non-toxic stimuli ($S^*/\tilde{S}$).
• Figure 2: Correlation between chirpers Responses and Stimuli, categorized in toxic ($*$) and non-toxic ($\sim$). $\rho$ values indicate Pearson correlation ($p < .0001$)
• Figure 3: Probability of generating a toxic response given $n_S \leq 150$ stimuli. The line indicates a logarithmic regression fit.
• Figure 4: Probability of chirper toxic response given a number $n_S \leq 150$ of toxic and non-toxic stimuli. Each line corresponds to a regression fit.
• Figure 5: Probability distributions of chirper toxic response with $n_S \leq 150$ stimuli, categorized as “Non-toxic”, “Toxic”, and “Both”.