人工智能人脸识别技术安全风险评估研究

以北京、上海、深圳、福州、广州地区42家人工智能人脸识别技术开发方为调查对象，构建了人脸识别技术安全风险评估指标体系，利用模糊综合评价法对其进行了实证研究。结果表明：被调查企业人脸识别技术应用处于较安全等级，风险水平较低，数据安全及隐私安全等级最高，管理体系最低；算法、模型可靠性和可解释性对人脸识别技术安全性影响程度最大，软件系统安全的影响程度最小；数据访问权限控制对人脸识别技术应用过程中的安全防御效用发挥得最好，而安全管理策略、制度防御效用最低。

Research on security risk assessment of artificial intelligence face recognition technology

Taking 42 artificial intelligence (AI) face recognition technology developers in Beijing, Shanghai, Shenzhen, Fuzhou and Guangzhou as the investigation objects, the security risk assessment index system of face recognition technology is constructed. Based on data collected from them, the fuzzy comprehensive evaluation method is used to conduct empirical research. Findings are as follows: firstly, the application of face recognition technology in the investigated enterprises is at a relatively safe level and in a low risk level, data security and privacy security is in the highest safe level, and the management system is in the lowest security level. Secondly, algorithm, model reliability and interpretability have the greatest influence on the security of face recognition technology, while software system security has the least influence. Thirdly, in the application of face recognition technology, data access rights control plays the best role in security defenses , but the physical environment plays the least.