paper-with-me

Papers

SafetyAnalyst: Interpretable, Transparent, and Steerable Safety Moderation for AI Behavior

2024-10-22 · Jing-Jing Li, Valentina Pyatkin, Max Kleiman-Weiner, Liwei Jiang, Nouha Dziri, Anne G. E. Collins, Jana Schaich Borg, Maarten Sap, Yejin Choi, Sydney Levine

The ideal AI safety moderation system would be both structurally interpretable (so its decisions can be reliably explained) and steerable (to align to safety standards and reflect a community's values), which current systems fall short on. To address this gap, we present SafetyAnalyst, a novel AI safety moderation framework. Given an AI behavior, SafetyAnalyst uses chain-of-thought reasoning to analyze its potential consequences by creating a structured "harm-benefit tree," which enumerates harmful and beneficial actions and effects the AI behavior may lead to, along with likelihood, severity, and immediacy labels that describe potential impacts on stakeholders. SafetyAnalyst then aggregates all effects into a harmfulness score using 28 fully interpretable weight parameters, which can be aligned to particular safety preferences. We applied this framework to develop an open-source LLM prompt safety classification system, distilled from 18.5 million harm-benefit features generated by frontier LLMs on 19k prompts. On comprehensive benchmarks, we show that SafetyAnalyst (average F1=0.81) outperforms existing moderation systems (average F1$<$0.72) on prompt safety classification, while offering the additional advantages of interpretability, transparency, and steerability.

📄 PDF Abstract BibTeX arXiv:2410.16665

Code (0)

등록된 구현이 없습니다.

Tasks

Knowledge Distillation

Methods 이 논문이 사용한 방법론

ALIGN In the ALIGN method, visual and language representations are jointly trained from noisy image alt-text data. The image and text encoders are learned via contrastive loss…
SET Dynamic Sparse Training method where weight mask is updated randomly periodically

Similar Papers 제목 키워드 기반

Aetheria: A multimodal interpretable content safety framework based on multi-agent debate and collaboration

2025-12-02 · Yuxiang He, Jian Zhao, Yuchen Yuan, Tianle Zhang 외 arxiv

The exponential growth of digital content presents significant challenges for content safety. Current moderation systems, often based on single models or fixed pipelines, exhibit limitations in identifying implicit risks…

Agentic Moderation: Multi-Agent Design for Safer Vision-Language Models

2025-10-29 · Juan Ren, Mark Dras, Usman Naseem arxiv

Agentic methods have emerged as a powerful and autonomous paradigm that enhances reasoning, collaboration, and adaptive control, enabling systems to coordinate and independently solve complex tasks. We extend this paradi…

Decoding the Rule Book: Extracting Hidden Moderation Criteria from Reddit Communities

2025-09-03 · Youngwoo Kim, Himanshu Beniwal, Steven L. Johnson, Thomas Hartvigsen arxiv

Effective content moderation systems require explicit classification criteria, yet online communities like subreddits often operate with diverse, implicit standards. This work introduces a novel approach to identify and …

UNIVID: Unified Vision-Language Model for Video Moderation

2026-06-04 · Kejuan Yang, Yizhuo Zhang, Mingyuan Du, Yue Zhang 외 arxiv

Global-scale video moderation faces a dual challenge: the need for fine-grained multi-modal reasoning and the demand for interpretable outputs to support downstream enforcement. Traditional moderation systems often rely …

TANDEM: Temporal-Aware Neural Detection for Multimodal Hate Speech

2026-01-16 · Girish A. Koushik, Helen Treharne, Diptesh Kanojia arxiv

Social media platforms are increasingly dominated by long-form multimodal content, where harmful narratives are constructed through a complex interplay of audio, visual, and textual cues. While automated systems can flag…

Reinforcement LearningBinary ClassificationTemporal Sequences