LLM危害分类法：五阶段梳理大模型风险
This study builds a simple map of all the main ways large language models can ca…
This study builds a simple map of all the main ways large language models can cause harm.
这项研究构建了一张简单的地图，展示了大型语言模型可能造成危害的所有主要方式。
Using about 200 research papers and incident reports, it groups harms into 5 stages of a model's life.
利用约200篇研究论文和事件报告，它将危害按模型生命周期的5个阶段进行分组。
Before the model is released, it highlights issues like scraped personal data, training on text without people's consent, heavy energy use, and low paid annotation work.
在模型发布前，它强调了诸如抓取个人数据、未经同意使用文本进行训练、高能耗以及低薪标注工作等问题。
In what the model writes out, it focuses on biased or stereotyped language, toxic or false content, and hallucinations that look trustworthy.
在模型输出的内容中，它聚焦于有偏见或刻板印象的语言、有毒或虚假内容，以及看似可信的幻觉。
For intentional misuse, it shows how people can generate scams, targeted abuse, propaganda, and even prompt based attacks on connected systems.
对于故意滥用，它展示了人们如何生成诈骗、针对性辱骂、宣传，甚至对连接系统进行基于提示的攻击。
At the broader social level, it describes job disruption, political manipulation, concentration of computing power, and unequal access to advanced models across regions and languages.
在更广泛的社会层面，它描述了就业干扰、政治操纵、计算能力集中，以及不同地区和语言间获取先进模型的不平等。
When models are built into tools for healthcare, finance, education, or creative work, it explains how errors and bias can quietly shape real decisions.
当模型被整合到医疗、金融、教育或创意工作的工具中时，它解释了错误和偏见如何悄然影响现实决策。
Across all stages, the authors line up existing technical and policy defenses and argue that only many layered safeguards together can keep risks manageable.
在所有阶段中，作者梳理了现有的技术和政策防御措施，并认为只有多层防护措施共同作用，才能将风险控制在可控范围内。
– arxiv. org/abs/2512.05929
– arxiv.org/abs/2512.05929
Paper Title: "LLM Harms: A Taxonomy and Discussion"
论文标题：《LLM危害：分类与讨论》
更进一步：量化金融体系
看懂新闻只是起点——沿量化金融路径，把它变成能交付的工程能力