Reproduce “UniGuard: Multi-View Guardrail Learning for Cross-Interface Safety Generalization” 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

Viet-Thanh Pham, Thuy-Trang Vu, Junae Kim, Rollin Omari, Dinh Phung, Gholamreza Haffari, Lizhen Qu

UniGuard trains a guardrail to classify safety cases across prompts, responses, conversations, agent traces, and multi-agent traces. It aligns views of the same case through supervised contrastive learning and uses cross-view agreement to calibrate confidence. Across ATBench, PolicyGuardBench, and DynaBench, UniGuard-0.6B gains 45.12 F1 points and reduces ECE by 35.41 points versus Qwen3Guard-Gen-0.6B on studied schema and policy shifts, not arbitrary safety domains.

Summary by alphaXiv.

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