Reproduce “Black-Box Red-Teaming of Multi-Agent Systems via Reinforcement Learning” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Yuqi Jia, Hangsheng Zhang, Chenxi Liu, Junjie Liang, Xun Chen

ReMAS targets system-prompt extraction in multi-agent systems by learning two policies: one rewrites an attack prompt, and another wraps it to reach a specialist while preserving its content. The policies run once each at test time. In a simulated Qwen3-8B benchmark, ReMAS achieved 0.99 CAR and 0.89 ASR for Coding, versus 0.52 and 0.16 for AutoDAN-Turbo; deployed-system performance remains unestablished.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.