《保险研究》20250708-《论人工智能核保风险的保险监管》（吴秀、于海纯）

［摘   要］人工核保向人工智能核保转变的两大突出表现是数据收集从“投保人如实告知”到“让替代性核保数据说话”的转变，以及数据评估从“数理查定法”到“机器学习算法”的转变，前者可能引致风险评估失真、核保偏见和金融排斥，后者可能加剧前述风险并引致模型风险、保险信任危机和金融稳定风险。因应人工智能核保的风险生成型态，应确立比例原则、协同共治原则、利益平衡原则为规制人工智能核保风险应遵循的保险监管原则。在监管规则层面，首先应规范替代性核保数据的使用，明确替代性核保数据的引入必须同时满足精算逻辑和价值判断逻辑；其次应加强对人工智能核保算法的规制，确立“前端认证+后端审计”的规制模式；最后还应扩张保险人的信息披露义务，明确保险人应通过公司官方网站披露等集约化方式履行使用人工智能核保系统的一般信息披露义务、通过核保通知书等特别方式履行针对特定投保人核保结论的具体信息披露义务。

［关键词］人工智能；核保；风险控制；保险监管；监管原则；监管规则

［基金项目］本文受对外经济贸易大学研究生科研创新基金资助。

［作者简介］吴秀，对外经济贸易大学法学院博士研究生；于海纯，对外经济贸易大学深圳研究院院长，法学院教授、博士生导师，研究方向：商法学。

Insurance Regulation on AI-Powered Underwriting Risks

WU Xiu，YU Hai-chun

Abstract：The two prominent shifts in the transition from manual underwriting to AI-powered underwriting are the change in data collection from "disclosure by the insurance applicant" to "letting alternative underwriting data speak"，and the change in data evaluation from the "numerical rating method" to "machine learning algorithms".The former may lead to distorted risk assessments，underwriting bias，and financial exclusion，while the latter may exacerbate these risks and result in model risks，insurance trust crises，and financial stability risks.In response to the risk generation patterns of AI-powered underwriting，the regulatory principles to address AI underwriting risks should be based on the proportionality principle，the principle of collaborative governance，and the principle of balancing interests.At the regulatory rule level，the following measures should be adopted：firstly，the use of alternative underwriting data should be regulated，ensuring that the introduction of such data meets both actuarial logic and value judgment logic;secondly，stronger regulation on AI underwriting algorithms should be implemented，establishing a "front-end certification and back-end auditing" regulatory model;finally，the insurer’s information disclosure obligation should be expanded.Insurers should fulfill their general information disclosure obligation regarding the use of AI-powered underwriting systems through centralized means such as official websites，and their specific disclosure obligation regarding underwriting decisions for individual insurance applicants through targeted means such as underwriting notices.

Key words：artificial intelligence;underwriting;risk control;insurance regulation;regulatory principles;regulatory rules