档案安全保障2026年第40卷第2期《档案学研究》
大语言模型驱动的档案开放审核安全风险防控机制构建研究
Research on the Construction of Security Risk Prevention and Control Mechanism for Archival Open Review Driven by Large Language Model
盐城师范学院历史与公共管理学院/法学院 盐城 224007
摘要
在档案管理数字化与智能化进程中，大语言模型（LLM）的应用在提升档案审核效率的同时，因数据敏感性、模型缺陷等引发安全风险，传统技术与管理模式难以应对。本研究基于档案学“信息生命周期理论”与“可解释人工智能”（XAI）理论，通过剖析 LLM 与档案审核业务的耦合机理，从数据层、模型层、应用层系统识别安全风险，构建“数据全生命周期防护—模型安全加固—智能审核流程优化”的技术防控体系，并配套设计审核安全管理制度、合规伦理治理框架及可信评估监督机制，形成“技术—管理—法律”协同的安全风险防控模式，为档案开放审核的智能化、安全化发展提供支撑，推动档案事业在技术赋能下实现安全与效率的协同提升。
关键词：大语言模型档案开放审核安全风险数据防护模型安全
Abstract
In the process of digitalization and intelligentization of archival management, the application of Large Language Model (LLM) has improved the efficiency of archival review. However, it has also given rise to security risks due to data sensitivity and model flaws, which are difficult for traditional technologies and management models to address. Based on the "Information Lifecycle Theory" in archival science and the "Explainable Artificial Intelligence" (XAI) theory, this study analyzes the coupling mechanism between LLM and archival review services, systematically identifies security risks at the data, model and application layers and constructs a system featuring "full-lifecycle data protection, model security enhancement, and intelligent review process optimization". In addition, it designs supporting review security management systems, compliant ethical governance frameworks, and credible evaluation and supervision mechanisms, forming a collaborative "technology-management-law" security risk prevention and control model. This model provides support for the intelligent and secure development of archival open review, and promotes the coordinated improvement of security and efficiency in the archival industry through technical empowerment.
Key words: Large Language Model; archival open review; security risk; data protection; model security
引用格式
参考文献
