Distributed Quality-Diversity Search for Toxicity in Large Language Models

2026-06-23

AI Summary

This work addresses the vulnerability of large language models to adversarial prompts that elicit harmful outputs, a challenge exacerbated by the high computational cost of existing red-teaming methods, which limits their ability to explore diverse failure modes. To overcome this, the authors propose ToxSearch-S, an evolutionary algorithm–based approach for toxicity-optimized prompt search. ToxSearch-S introduces an embedding-driven incremental speciation mechanism to preserve behavioral diversity and integrates an MPI master-worker architecture for efficient distributed parallel search. Experimental results demonstrate that, under identical computational budgets, ToxSearch-S achieves peak toxicity levels comparable to those of ToxSearch and RainbowPlus but with lower cumulative toxicity. Moreover, it attains a 3.2× speedup on a 4-node configuration while significantly increasing both the final number of species and the proportion of toxic species.

Abstract

Large Language Models remain vulnerable to adversarial prompts that elicit harmful responses, and scaling red-teaming to cover a broad range of failure modes is constrained by the cost of text generation and evaluation. We present \emph{ToxSearch-S}, a speciated extension of toxicity-focused evolutionary prompt search with incremental, embedding-driven niche maintenance, together with an MPI master-worker realization that centralizes population and species bookkeeping on rank~0 while offloading prompt evolution and evaluation to $n_w$ parallel workers. Under a common budget, ToxSearch-S attains peak toxicity competitive with both ToxSearch and RainbowPlus while following a measurably less toxic best-so-far trajectory, indicating lower cumulative search pressure. Diversity is non-uni-dimensional: RainbowPlus yields greater embedding-level spread, whereas ToxSearch-S partitions high-toxicity prompts into more localized behavioral pockets, reflected by a higher DBSCAN cluster count. MPI distribution delivers substantial wall-clock gains, approximately $1.8\times$ with two workers and $3.2\times$ with four, while leaving Best@B statistically indistinguishable from sequential execution. Four-worker runs also produce significantly larger final species cardinality and more toxicity-bearing species, without a reliable gain in global peak toxicity. These results position incremental speciation as a practical quality-diversity mechanism for AI Safety and MPI as an effective means of compressing time-to-result while preserving measured search outcomes.

Problem

• toxicity
• adversarial prompts
• red-teaming
• large language models
• quality-diversity search

Innovation

• speciation
• quality-diversity
• distributed evolutionary search
• MPI parallelization
• toxicity red-teaming