ProGuard: Proactive AI Safety

Published:Dec 29, 2025 16:13
1 min read
ArXiv

Analysis

This paper introduces ProGuard, a novel approach to proactively identify and describe multimodal safety risks in generative models. It addresses the limitations of reactive safety methods by using reinforcement learning and a specifically designed dataset to detect out-of-distribution (OOD) safety issues. The focus on proactive moderation and OOD risk detection is a significant contribution to the field of AI safety.

Reference

ProGuard delivers a strong proactive moderation ability, improving OOD risk detection by 52.6% and OOD risk description by 64.8%.