Year: 2026
Type: conference-paper
Abstract: Warning: This paper contains content that may be inappropriate, offensive, or harmful.Automated red-teaming has emerged as a scalable approach for auditing Large Language Models (LLMs) prior to deployment, yet existing approaches lack mechanisms to efficiently adapt to model-specific vulnerabilities at inference.We introduce Red-Bandit, a red-teaming framework that adapts online to identify and exploit model failure modes under distinct attack styles (e.g., manipulation, slang).Red-Bandit post-trains a set of parameter-efficient LoRA experts, each specialized for a particular attack style, using reinforcement learning that rewards the generation of unsafe prompts via a rule-based safety model.At inference, a multiarmed bandit policy dynamically selects among these attack-style experts based on the target model's response safety, balancing exploration and exploitation.Red-Bandit outperforms stateof-the-art methods on AdvBench and Harm-Bench, achieving higher attack success rates under sufficient exploration budgets (ASR@10), while generating more human-readable adversarial prompts (lower perplexity).In addition, Red-Bandit's bandit policy serves as a diagnostic tool for identifying model-specific vulnerabilities by indicating which attack styles most effectively elicit harmful behaviors.
Authors: Christos Ziakas, Nicholas Loo, Nishita Jain, Alessandra Russo
Institution: Imperial College London
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W4415318193