Reproduce “GuardrailAgent: An Agentic Framework for Jailbreak Defense” 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
Tianze Chang, Junjie Chu, Yixin Wu, Michael Backes, Yang Zhang
GuardrailAgent combines prompt-side and response-side jailbreak checks with a knowledge base that can be updated when an input miss is caught at the output. In a feedback-only experiment, its input detection rate rose from 0.033 to 0.800 after about 90 adversarial prompts across three runs. This benchmark result does not establish protection against all future attacks.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.