论文/arXiv 2610.09082·22 天前

快速行动，及时修补：利用社交媒体评论跟上不断演变的 AI 危害

提出 LLM 辅助主题分析流程，从大规模 Reddit 数据动态检测、分类和追踪新兴 AI 危害。

AI 整理

• 数据与采集
• 具身模型
• 端侧部署

原文链接：Move Fast and Mend Things: Keeping Up with Evolving AI Harms Using Social Media Commentary（arXiv 2610.09082 · 发布日期：2026-10-06）

Move Fast and Mend Things: Keeping Up with Evolving AI Harms Using Social Media Commentary

提出 LLM 辅助主题分析流程，从大规模 Reddit 数据动态检测、分类和追踪新兴 AI 危害。

核心要点

• 动机是快速部署的 AI 系统会产生难以被事前建模和事后事件追踪捕捉的危害。
• 方法用 LLM 辅助主题分析流程，从大规模社交媒体数据中动态发现和组织 AI 危害。
• 在 18 个月 570 万条 Reddit 帖子摘要上应用，形成 57.5 万条相关帖子数据集。
• 发布自底向上的 AI 危害分类法，含 12 类和 47 个子节点，并补充参与式治理。

验证范围与局限：验证范围为 2025 年 1 月至 2026 年 6 月的 Reddit 帖子摘要分析。

英文摘要 (Abstract)

The rapid deployment of AI systems has created socio-technical, psychological, and operational harms that can elude ex-ante threat modelling and ex-post incident tracking. We introduce an LLM-assisted thematic analysis pipeline to dynamically detect, categorise, and track emerging AI harms from large-scale social media data. Applying it to 5.7 million Reddit post summaries over 18 months (01/2025 to 06/2026), we curate and release a dataset of 575,000 AI harm-related posts and a bottom-up AI harm taxonomy of 12 categories and 47 subnodes. The taxonomy reliably covers established expert-defined risks while surfacing granular harms that top-down frameworks overlook, such as distinct forms of AI privacy violations. Temporal analysis surfaces evolving user-centric harms, such as agentic privacy and security breaches, premature AI adoption in the workplace, and grief from AI companion discontinuation. Our pipeline shortens harm-detection timelines and hereby complements efforts towards more participatory and responsive AI governance.

作者：Jacqueline Rowe, Animesh Srivastava, Sai Teja Peddinti, Seliem El-Sayed, Nina Taft

来源

本页事实对应的公开页面，以原文为准。

1. arXiv 2610.09082：Move Fast and Mend Things: Keeping Up with Evolving AI Harms Using Social Media Commentary访问 2026-10-21