您正在使用IE低版浏览器，为了您的雷峰网账号安全和更好的产品体验，强烈建议使用更快更安全的浏览器
此为临时链接，仅用于文章预览，将在时失效
0
导语：11月3日，香港大学在2026年香港金融科技周发布全球首个KYA报告《KYA：智能体信用体系研究报告》，提出智能体成熟度模型、KYA框架与发展指数。
11月3日，香港大学副校长汪扬教授与香港大学经管学院杨宏副教授代表香港大学KYA课题组，在2026年香港金融科技周 x StartmeupHK创业节发布全球首个KYA（Know Your Agent）领域报告《KYA：智能体信用体系研究报告》。
报告聚焦智能体参与真实经济活动后的身份、授权、信用与责任问题，结合109份从业者有效问卷及包含Stripe、蚂蚁数科等近十个企业案例，提出智能体成熟度模型、KYA框架和智能体发展指数，并进一步探讨“智能体信任资本”的形成及其经济价值。
“没有KYA，就没有智能体经济。”报告认为，模型能力的提升不会自动转化为商业上的可委托性。只有当身份可确权、履约可核验、权限可约束时，智能体才能获得与其能力相匹配的经济授权。KYA所要解决的，正是智能体从技术可能走向商业可行的问题。
香港大学副校长汪扬教授与香港大学经管学院杨宏副教授发布全球首个KYA领域报告
智能体加速进入真实经济，信任赤字正侵蚀人工智能红利
报告将智能体的演进概括为四个阶段：能力智能体、执行智能体、交易智能体和经济智能体。截至目前，行业整体仍处于从能力智能体向执行智能体加速演进的阶段，部分前沿场景已开始探索交易智能体。
从执行任务走向跨主体交易，意味着智能体需要获得企业内部流程之外的信任。采购、支付或专业服务智能体可以在授权范围内代表用户采取行动，但交易对手仍需确认其代表谁、获得何种授权、能否可靠履约，以及出现损失时由谁承担责任。
行业研究显示，智能体的企业部署意愿正在快速上升。Gartner在2026年《Agentic AI技术成熟度曲线》中指出，受访组织中已有17%部署AI智能体，超过60%预计在未来两年内部署。Gartner另预测，到2028年，全球《财富》500强企业平均使用的智能体数量将超过15万个。
智能体商业活动的潜在规模也受到关注。麦肯锡估计，到2030年，智能体商业可能为全球零售市场带来3万亿至5万亿美元的协调收入机会；该估算目前仅涉及商品零售，不包括服务和B2B市场。
与能力快速提升形成鲜明对照的是，支撑智能体运行的信任体系几乎仍是空白。报告将这一能力与制度之间的错位概括为“智能体信任真空”：过去两年，智能体越权操作、提示词注入、工具投毒、多智能体协作责任失效等安全事件密集出现，并直接转化为可计量的经济成本。
Cloud Security Alliance与Anjuna对全球金融业的调查显示，62%的受访机构已部署AI智能体。而Accenture研究发现，87%的金融机构受访者认为，“信任”将成为智能体支付规模化应用的最大障碍。
杨宏在分享中指出：“智能体正在走进真实经济，其扩散速度远远快于互联网发展期的普及速度。但是，信任赤字，正在侵蚀人工智能红利。”
76%受访企业存在智能体信任顾虑，信任摩擦成为商业隐性成本
行业问卷显示，智能体信任摩擦已经常态化：76%的受访从业者在过去一年频繁或偶尔遇到与智能体信任、安全、问责或合规有关的顾虑，其中23%频繁遇到，53%偶尔遇到，仅2%表示从未遇到。
信任顾虑也已产生实际经营影响：41%的受访者提及合规与审计压力上升，40%提及项目上线或扩展延迟，38%提及人工确认增加、效率下降，23%表示曾遭遇客户或合作伙伴拒绝使用。
在推进智能体商业应用的主要障碍中，“客户或伙伴信任度低”（53%）和“出错后责任归属不清”（42%）位列前两项，高于系统集成困难（35%）与投资回报难以测算（34%）。
香港大学《KYA：智能体信用体系研究报告》主要发现
需求侧共识同样强烈。88%的受访者认为，行业需要统一的智能体可信评估或认证标准，60%愿意主动采用相关服务，或先在高风险、关键场景中试点。
课题组认为，智能体经济发展的瓶颈正在从单纯的技术能力，转向市场能否判断并管理智能体的行动风险。没有可信身份、清晰授权、可核验履约记录和责任安排，企业与消费者就难以放心扩大委托，智能体也难以从内部任务执行者走向跨主体交易参与者。KYA因此成为连接智能体技术与真实经济活动的核心枢纽。
“智能体经济的主要约束，正在由技术能力转向信任体系。”汪扬强调，“当智能体进入支付之后，信任问题就直接转化为商业问题。KYA的目标，是安全地释放智能体经济的规模。”
“信任资本”将影响智能体的商业价值与竞争优势
技术能力相近的智能体，可能具有不同的商业价值。报告认为，可信身份、明确授权、可靠履约记录及责任保障，会影响企业是否愿意委托，以及愿意赋予多大权限。
报告据此提出“智能体信任资本”概念：经济主体通过可信治理投入和智能体持续、可验证的履约活动，形成能够降低交易成本、扩大经济活动范围并创造未来收益的无形资本。
随着可信记录积累，合作方可以减少重复审查，企业有机会扩大业务委托，提高既有AI能力的利用效率和投资回报。
蚂蚁数科智能体超级工厂是报告收录的典型案例之一。其Agentar平台覆盖智能体开发、部署、运行与治理。依托区块链、数字身份、隐私计算和安全风控的长期布局，蚂蚁数科智能体超级工厂已将身份核验、授权管理及行为记录纳入业务流程。报告认为，智能体平台有机会将可信能力嵌入整个生命周期，为信任资本的持续积累提供基础。
据悉，香港金融科技周期间，蚂蚁数科发布KYA产品升级，为其智能体超级工厂模式下全线产品提供身份基础设施。
2026年香港金融科技周 x StartmeupHK创业节期间，蚂蚁数科展示KYA产品升级
此次报告收录九类正在建立智能体信任的代表企业，覆盖支付、金融、教育、能源、医疗等领域。报告展望，产业将迎来四类战略机遇：生产智能体——从销售软件工具转向提供数字劳动力；组织智能体——从管理软件流程转向组织和配置智能体生产力；信任智能体——从降低单笔交易风险，发展为提供智能体信用服务、培育信任资本；连接智能体——从服务企业数字化转型，进一步连接跨平台的智能体服务与经济活动。
“人工智能时代的经济竞争，将不仅是模型、算力和算法之间的竞争，也将日益体现为生产能力的可信程度、组织效率以及经济协作范围之间的竞争。”汪扬表示。
雷峰网(公众号：雷峰网) 雷峰网
为了您的账户安全，请验证邮箱
请验证您的邮箱
本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
You are using a low-end IE browser and strongly recommend a faster and safer browser for your Thunder Peak account security and better product experience.
This is a temporary link, only for article preview, which will expire at the time
0
Introduction: On 3 November, the University of Hong Kong presented the AI agent maturity model, the KYA framework and development index for the first KYA study on credit systems in the Hong Kong Financial Science and Technology Week 2026 publish worldwide.
On 3 November, Professor Wang Yang, Vice-Chancellor of the University of Hong Kong and Associate Professor Yang Hong Kong, on behalf of the KYA Group at the University of Hong Kong, reported on the study of the KYA:AI agent credit system in the first region of the ZQYA (Know Your Agent) worldwide, XStartmeupHK Entrepreneurship Festival, 2026.
The report focuses on identity, empowerment, credit and responsibility following the involvement of AI agent in real-life economic activities, and presents the AI agent maturity model, the KYA framework and the AI agent development index, together with 109 valid questionnaires for practitioners and nearly 10 business cases including Stripe, Ant Numbers Section, and further explores the formation of “AI agent trust capital” and its economic value.
“Without KYA, there is no AI agent economy.” The report concludes that the upgrading of modelling capacity does not automatically translate into commercial commissionability. AI agent can only obtain an economic authorization matching its capabilities if it is identifiable, performance verifiable and permission binding. It is precisely AI agent that QYA has to address the issue of the possibility of technology moving from commercial to commercial.
Report by Professor Wang Yang, Vice-Rector, University of Hong Kong, and Associate Professor Yang Hong, University of Hong Kong School of Administration, publish, the first KYA area worldwide
AI agent accelerates into the real economy, trust deficits erode artificial intelligence dividends
The report summarizes the evolution of AI agent in four stages: capability AI agent, implementation of AI agent, trading AI agent and economic AI agent. To date, the industry as a whole is still in the process of accelerating the evolution from capability AI agent to the implementation of AI agent, and some of the frontier scenes have begun to explore trading AI agent.
Moving from mandate implementation to cross-curricular transactions means that AI agent needs to gain confidence outside the internal processes of the enterprise. Procurement, payment or professional services AI agent may act on behalf of the user within the delegated authority, but the counterparty still needs to confirm who it represents, what authority it receives, whether it can reliably perform, and who will be held liable in the event of loss.
Industry studies show that the willingness of AI agent enterprises to deploy is rising rapidly. Gartner noted in the 2026 Agenic AI technological maturity curve that 17 per cent of the organizations visited had deployed AIAI agent, and over 60 per cent were expected to deploy within the next two years. Gartner also predicts that the average number of AI agents used by the Global Wealth 500 companies will exceed 150,000 by 2028.
The potential scale of AI agent ' s commercial activities was also noted. McKinsey estimates that by 2030, AI agent businesses could generate US$ 3 trillion to US$ 5 trillion in coordinated income opportunities for the global retail market; this estimate currently covers only retail trade in commodities, excluding services and B2B markets.
In contrast to the rapid increase in capacity, the trust system that underpins the operation of AI agent remains almost blank. The report summarizes the mismatch between this capability and the system as “AI agent confidence vacuum”: over the past two years, security incidents such as AI agent ultra vires operations, prompt injections, tool poisoning, multi AI agent failure of responsibilities for collaboration have been intense and directly translated into measurable economic costs.
A survey of the global financial sector conducted by Claude Security Alliance and Anjuna showed that 62 per cent of the institutions interviewed had deployed AIAI agent. The Accenture study found that 87 per cent of the respondents to financial institutions believed that “trust” would be the biggest obstacle to the scale-up of AI agent payments.
In his sharing, Yang Hong stated: “AI agent is entering the real economy and is spreading much faster than the spread of the Internet. But the trust deficit is eroding artificial intelligence dividends.”
76 per cent of the enterprises interviewed had AI agent confidence concerns and trust friction became a hidden commercial cost
The industry questionnaire shows that the friction of AI agent confidence has become regular: 76 per cent of the respondents have experienced frequent or occasional concerns over AI agent trust, security, accountability or compliance over the past year, 23 per cent of which were encountered frequently, 53 per cent occasionally and only 2 per cent indicated never.
Confidence concerns have also had an operational impact: 41 per cent of respondents referred to an increase in compliance and audit pressure, 40 per cent to delays in project start-up or expansion, 38 per cent to an increase in human confirmation and a decrease in efficiency, and 23 per cent to a refusal by clients or partners to use it.
Among the major obstacles to the advancement of AI agent commercial applications, “low customer or partner confidence” (53 per cent) and “ununderstanding attribution of responsibility after error” (42 per cent) rank first, higher than the difficulties of system integration (35 per cent) and investment returns (34 per cent).
KYA:AI agent Research on Credit Systems, University of Hong Kong
Demand-side consensus is equally strong. Eighty-eight per cent of respondents were of the view that the industry needed a unified AI agent credible assessment or certification standard, and 60 per cent were willing to take the services on their own initiative or pilot them in high-risk, critical scenarios.
The group considered that the bottlenecks in AI agent ' s economic development were shifting from purely technical capacity to the ability of the market to judge and manage the operational risks of AI agent. Without credible identities, clear authorizations, verifiable performance records and accountability arrangements, it is difficult for businesses and consumers to secure the extension of delegation, and it is difficult for AI agent to move from internal mandate holders to cross-cutting players. KYA has thus become a central hub linking AI agent technology to real economic activity.
“The main constraint of the AI agent economy is shifting from technological capability to a system of trust.” Wang Ying stressed that “when AI agent enters the payment, the issue of trust becomes a business issue directly. The goal of KYA is to release safely the size of the AI agent economy.”
Trust capital will affect the commercial value and competitive advantage of AI agent
AI agent, which has similar technological capabilities, may have different commercial value. The report considers that credible identity, clear delegation of authority, reliable record of performance and guarantees of liability affect the willingness of enterprises to commission, and the extent to which they are willing to grant permission.
The report thus introduces the concept of “AI agent trust capital”: economic agents form intangible capital that reduces transaction costs, expands the scope of economic activity and creates future gains through credible governance inputs and continuous and verifiable compliance activities by AI agent.
As credible records accumulate, cooperation can reduce duplication of reviews, and enterprises have the opportunity to expand their business assignments and increase efficiency in the use of existing AI capabilities and return on investment.
One of the typical cases reported was the AI agent Super Factory of the Ant Numbers Section. Its Agental platform covers the development, deployment, operation and governance of AI agent. In the long-term layout of the ants block chain, digital identity, privacy computing and safety wind control, the ants section AI agent supershop has incorporated identification, authorization management and behavioural records into business processes. The report found that the AI agent platform had the opportunity to embed credible capabilities throughout the life cycle, providing a basis for the continued accumulation of trust capital.
According to the information received, during the Hong Kong Financial Science and Technology Week, the ant count section publishKYA product was upgraded to provide a full-line identification infrastructure for its AI agent super factory model.
During the Business Festival of the Financial Technology Week in Hong Kong in 2026, X StartmeupHK, the Ant Numbers Section demonstrated the upgrading of the KYA product
This report includes nine categories of representative enterprises that are in the process of establishing AI agent trust, covering areas such as payment, finance, education, energy and health care. The report looks forward to four types of strategic opportunities for industry: production of AI agent - from marketing software tools to providing a digital workforce; organization AI agent - Moving from managing software processes to organizing and configuration AI agent productivity; trusting AI agent - From reducing the risk of single transactions to providing AI agent credit services and fostering trust capital; connecting AI agent - from the digitalization of service enterprises to further linking AI agent services across platforms to economic activities.
“Economic competition in the age of artificial intelligence will not only be a competition between models, algorithms and algorithms, but will increasingly be reflected in a competition between the credibility of productive capacity, organizational efficiency and the scope of economic collaboration.”
Thunder Peak Network.
For your account security, please verify the mailbox.
Check your mailbox.
Auto-fetched and translated by AI Atlas. All rights belong to the original author.