Beyond Static Benchmarks: Synthesizing Harmful Content via Persona-based Simulation for Robust Evaluation

TL;DR

This paper introduces a persona-guided LLM framework to synthesize diverse, harmful content for more effective and challenging evaluation of detection systems, overcoming static benchmark limitations.

Contribution

It presents a novel persona-based simulation framework that generates diverse, contextually grounded harmful content to improve robustness testing of detection models.

Findings

High harmful content generation success rate confirmed by human and LLM evaluations.

Synthetic scenarios are more challenging to detect than existing benchmarks.

Achieves linguistic and topical diversity comparable to human-curated datasets.

Abstract

Static benchmarks for harmful content detection face limitations in scalability and diversity, and may also be affected by contamination from web-scale pre-training corpora. To address these issues, we propose a framework for synthesizing harmful content, leveraging persona-guided large language model (LLM) agents. Our approach constructs two-dimensional user personas by integrating demographic identities and topical interests with situational harmful strategies, enabling the simulation of diverse and contextually grounded harmful interactions. We evaluate the framework along three dimensions: harmfulness, challenge level, and diversity. Both human and LLM-based evaluations confirm that our framework achieves a high harmful generation success rate. Experiments across multiple detection systems reveal that our synthetic scenarios are more challenging to detect than those in existing benchmarks. Furthermore, a multi-faceted analysis confirms that our approach achieves linguistic and topical diversity comparable to human-curated datasets, establishing our framework as an effective tool for robust stress-testing of harmful content detection systems.