---
title: "通过计算宣传投毒预训练数据：公共讨论界面作为攻击向量"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2607.15267v1"
permalink: "https://feed.astralor.com/items/itm_f720182eb4e172eedfed9e6a"
published: "2026-07-16T17:56:05.000Z"
score: 79
channel: "ai"
tags: ["arXiv","HalfLife","公共讨论接口","网络爬取","语言模型安全","预训练数据投毒"]
---

# 通过计算宣传投毒预训练数据：公共讨论界面作为攻击向量

## AI 点评

> 这篇论文值得所有关注LLM安全的人仔细阅读。它不满足于实验室里的“玩具攻击”，而是直接指向了互联网上每天都在发生的评论、论坛帖子——这些内容正是爬虫不断抓取的数据。对AI公司来说，这意味着数据清洗管线必须重新评估；对研究者而言，HalfLife框架提供了测量投毒留存率的实用工具。

## 速览

该论文展示了通过公共讨论界面（如论坛评论）向语言模型预训练数据中注入恶意内容的可行性，突破了以往依赖维基百科等固定数据源的研究局限。作者提出了HalfLife分析框架，用于估计网络爬取后对抗性内容在训练数据中的存留比例。实验表明，第三方网页内容可作为攻击LM预训练的潜在向量，且现有数据筛选流程难以完全防御此类投毒。

## 关键点

- 首次在真实网络规模下验证了通过公共讨论界面投毒预训练数据的可行性
- 提出HalfLife框架，用于量化对抗性内容在数据筛选后的存留比例
- 揭示了第三方网页内容作为LM预训练攻击向量的新风险
- 指出现有数据筛选流程对网络注入型投毒的防御不足

## 全文

Poisoning pretraining data can introduce harmful behaviors to LMs that are difficult to detect and mitigate. Prior work on poisoning pretraining data has largely exploited established data sources such as Wikipedia, which do not represent the large scale and heterogeneity typical of pretraining corpora, and has ignored the interaction between poisoned data and data curation pipelines. We demonstrate that poisoning attacks on pretraining data are feasible beyond this limited setting through an existing web-scale content injection mechanism: public discussion interfaces. Additionally, to measure whether malicious content is included after web crawling and data curation, we introduce HalfLife, a novel analysis for estimating adversarial content inclusion in web-crawl based LM training data. We use HalfLife to explore the feasibility of poisoning pretraining corpora at web scale through open discussion interfaces. Our analysis demonstrates the importance of estimating whether poison injections are included in pretraining data, and establishes third-party webpage content as a possible vector for attacking language model pretraining.

---

引用前请核对原文;本文由星际流动整理:https://feed.astralor.com/items/itm_f720182eb4e172eedfed9e6a