STAR-Teaming: A Strategy-Response Multiplex Network Approach to Automated LLM Red Teaming (ACL findings 2026).
While Large Language Models (LLMs) are widely used, they remain susceptible to jailbreak prompts that can elicit harmful or inappropriate responses. This paper introduces STAR-Teaming, a novel black-box framework for automated red teaming that effectively generates such prompts. STAR-Teaming integrates a Multi-Agent System (MAS) with a Strategy-Response Multiplex Network and employs network-driven optimization to sample effective attack strategies. This network-based approach recasts the intractable high-dimensional embedding space into a tractable structure, yielding two key advantages: it enhances the interpretability of the LLM's strategic vulnerabilities, and it streamlines the search for effective strategies by organizing the search space into semantic communities, thereby preventing redundant exploration. Empirical results demonstrate that STAR-Teaming significantly surpasses existing methods, achieving a higher attack success rate (ASR) at a lower computational cost. Extensive experiments validate the effectiveness and explainability of the Multiplex Network.
In line with responsible disclosure practices for red-teaming research, access to the codebase and attack artifacts is restricted to vetted researchers. This policy is intended to prevent misuse of automated jailbreak tooling while preserving the scientific value of the work. The code is provided strictly for non-commercial academic research purposes; any commercial use, malicious use, or redistribution is prohibited.
To request access, please contact us with:
- Your full name and institutional affiliation
- A brief description of your intended use
Requests are typically reviewed within 1–2 weeks.
For code access requests or research inquiries:
MinJae Jung — minjae.jung@selectstar.ai