GRA · 人工智能与金融模型风险
空间金融资本与危害、风险暴露和转型的交汇之处
行业范围与发布方
AI 与金融 模型风险 是全球风险联盟(Global Risks Alliance,GRA)的专业平台,位于其规范地址。其范围是模型治理以及在金融决策流程和风险计量中负责任地使用 AI。本页载明发布方(一家面向金融服务业的美国行业协会),介绍该平台所服务的专业群体,并说明适用于其全部工作的法律限制。
- STEM 方法
- 4
- 该领域应用的规范方法
- 主题
- 8
- 该领域的主题页面
- 对口方
- 5
- 网络中共享领域的各平台
- 报告系列
- 2
- 它参与的 Reports 系列
面向金融服务业的行业协会。
GRA 是在美国内华达州注册的非营利公司,作为金融服务业的行业协会运营。行业协会致力于整个行业的共同商业利益,而非任何单个成员的私人业务。GRA 是 Nexus Consortium 的三个机构之一,另外两个是全球风险与创新中心(Global Centre for Risk and Innovation,GCRI)和全球风险论坛(Global Risks Forum,GRF)。
The platform exists separately because finance-domain model risk differs from general technical AI safety and from generic fintech innovation. AI and Financial Model Risk is an established platform. Its pages set out its scope and how it is designed to work, and its activities, records and outputs are published here as each one opens. Information and enquiries are open today.
模型风险与物理、运营、技术和系统性风险交汇之处。
| 风险类别 | 渠道 | 框定 |
|---|---|---|
| 物理 | 驱动定价和资本的危害模型与巨灾模型;损害函数不确定的气候情景 | 巨灾模型与气候模型的模型验证 |
| 操作 | Model errors in production, poor change control, weak documentation | 模型风险 管理指南 |
| 技术 | 机器学习中的漂移、偏差和不透明性;生成式人工智能中的幻觉 | 欧盟《人工智能法案》;NIST《人工智能风险管理框架》 |
| 系统性 | Many institutions relying on the same model, data set or foundation model | Supervisory concern about common-mode exposure |
平台旨在产出的内容。
Documentation templates, validation method notes adapted for machine learning and studies of shared-model exposure, designed for the Methods and Technical series and the Exponential Technology series. Boundary. GRA validates, approves and recommends no model or vendor and gives no regulatory advice.
平台面向哪些人群。
The professional community spans banks, insurers, asset managers, market infrastructures and fintech firms, together with researchers who study model risk.
治理与监督
负责在整个机构内维护模型清单、政策和监督的人员。
独立质疑
Specialists who test model design, data, implementation and performance.
Ethics and accountability
Those responsible for fairness, explainability and human oversight of AI systems.
平台不做的事。
- 它不验证、批准、认证、评级或背书任何模型、人工智能系统、供应商或机构。
- It does not give legal, regulatory, investment or technical advice to any firm.
- 它不放贷、不承销、不评级、不投资,不安排或居间促成交易,不引荐交易对手,也不收取基于交易的报酬。
- Readiness work is not approval, and a published method is not a supervisory standard.
- Competitors cooperate only on precompetitive questions, under safeguards set out in a written competition policy. Proprietary models, pricing and customer data are not shared.
Decisions about any model remain with the institution that uses it and with its supervisors, under the law that applies to them.
Where this page’s references lead.
阅读本页时所对照的机构和文件。状态和版本会变化;每个链接都指向发布机构。
问题与解答。
AI and Financial Model Risk 是一个独立的组织吗?
供应商可以利用该平台推广模型吗?
合规问题应联系谁?
What is the platform’s current status?
法律文件。
本网站上的地图、数据和图表仅作为背景和一般信息展示。它们不是建议、预测或预警;信息的局限见免责声明和数据说明。