Diversifying Toxicity Search in Large Language Models Through Speciation

TL;DR

This paper introduces ToxSearch-S, a method that diversifies toxicity prompt search in large language models by maintaining multiple prompt niches, improving coverage of failure modes and toxicity levels.

Contribution

It presents a novel speciated quality-diversity approach that enhances toxicity search by preserving diverse prompt niches and exploring multiple failure modes simultaneously.

Findings

ToxSearch-S achieves higher peak toxicity than baseline.

Broader semantic coverage with higher topic diversity.

Prompt niches are well-separated in embedding space.

Abstract

Evolutionary prompt search is a practical black-box approach for red teaming large language models, however existing methods often collapse onto a small family of high-performing prompts, limiting coverage of distinct failure modes. We present a speciated quality-diversity extension of \textit{ToxSearch} that maintains multiple high-toxicity prompt niches in parallel rather than optimizing a single best prompt. \textit{ToxSearch-S} introduces unsupervised prompt speciation via a search methodology that maintains capacity-limited species with exemplar leaders, a reserve pool for emerging niches, and species-aware parent selection that trades off within-niche exploitation and cross-niche exploration. Preliminary results show \textit{ToxSearch-S} reaching higher peak toxicity ($\approx 0.73$ vs.\ $\approx 0.47$) with a heavier tail (top-10 median $0.66$ vs.\ $0.45$) than the baseline. Speciation also yields broader semantic coverage under a topics-as-species analysis (higher effective topic diversity and larger unique topic coverage). Finally, species formed are well-separated in embedding space (mean separation ratio $\approx 1.93$) and exhibit distinct toxicity distributions, indicating that speciation partitions the adversarial space into behaviorally differentiated niches rather than superficial lexical variants.