证券市场导报Issue(1)：3-16,14.
AI大模型驱动的智能博弈财务舞弊识别系统构建
深圳证券交易所财务舞弊监管AI大模型课题组 1郭旺振 1邓媛尹 1王佳奇 1张慧慧 1单益峰 1陈文新 2叶茂 3许明峰 4邱晓明 1兰茹婷 1黄锦鸿 1薛傲 1李铁照1
作者信息
- 1. 深圳证券交易所,广东 深圳 518038
- 2. 深圳证券交易所监管执行部
- 3. 深圳证券交易所创业板公司管理部
- 4. 深圳证券交易所信息科技二部
- 折叠
摘要
Abstract
The effective utilization of emerging technologies such as artificial intelligence to efficiently identify illegal violations and potential risks,and enhance the scientific rigor and effectiveness of capital market supervision,represents a crucial measure for implementing the requirements of the 15th Five-Year Plan proposal and the spirit of the Central Financial Work Conference.Based on the practice of digital and intelligent transformation of supervision at the Shenzhen Stock Exchange(SZSE),this paper thoroughly explores the application of large language models in identifying financial fraud issues.It innovatively proposes an intelligent fraud detection theoretical paradigm of"fraud detection chain-of-thought prompting+structured multi-dimensional information working papers+multi-agent game-theoretic confrontation",and develops a large language model-driven intelligent game-theoretic financial fraud detection system.This system specifically addresses current obstacles in applying large language models for financial fraud detection,effectively employing large language models to conduct"human-like"intelligent reasoning and analysis of financial fraud risks in listed companies,and provides regulatory personnel with alerts regarding potential fraud risks and regulatory response recommendations based on the analysis results.Relevant empirical testing results demonstrate that the system achieves high precision in fraud detection,with false negatives and false positives being well controlled.It effectively compensates for the shortcomings of machine learning models in fraud detection,as well as the problems of isolated fraud detection indicators and lack of comprehensive reasoning analysis under expert rule-based approaches,genuinely fulfilling its role as an intelligent assistant to regulatory personnel.The SZSE's construction of an AI large language model-driven intelligent game-theoretic financial fraud detection system represents an important exploration in responding to the State Council's"AI+"action initiative and promoting the digital and intelligent transformation of financial supervision.关键词
大模型/财务舞弊/舞弊识别思维链/多维信息工作底稿/智能体/博弈对抗Key words
large language models/financial fraud/chain-of-thought for fraud detection/multi-dimensional information working papers/intelligent agents/game-theoretic confrontation分类
管理科学引用本文复制引用
深圳证券交易所财务舞弊监管AI大模型课题组,郭旺振,邓媛尹,王佳奇,张慧慧,单益峰,陈文新,叶茂,许明峰,邱晓明,兰茹婷,黄锦鸿,薛傲,李铁照..AI大模型驱动的智能博弈财务舞弊识别系统构建[J].证券市场导报,2026,(1):3-16,14.
