University of Hong Kong Releases World's First KYA Report: Agent Trust Deficit Is Eroding AI Dividends

On November 3, Professor Wang Yang, Vice-President of the University of Hong Kong, and Associate Professor Yang Hong of the HKU Business School, representing the HKU KYA research team, released the world's first report in the KYA (Know Your Agent) field, titled "KYA: Research Report on the Agent Credit System," at the 2026 Hong Kong FinTech Week x StartmeupHK Festival.

The report focuses on the issues of identity, authorization, credit, and liability that arise after agents participate in real economic activities. Drawing on 109 valid practitioner questionnaires and nearly ten corporate case studies, including Stripe and Ant Digital Technologies, it proposes an agent maturity model, a KYA framework, and an agent development index, and further explores the formation of "agent trust capital" and its economic value.

"Without KYA, there is no agent economy." The report argues that improvements in model capabilities will not automatically translate into commercial delegability. Only when identity can be authenticated, performance can be verified, and permissions can be constrained can agents obtain economic authorization commensurate with their capabilities. What KYA seeks to solve is precisely the problem of agents moving from technical possibility to commercial viability.

Professor Wang Yang, Vice-President of the University of Hong Kong, and Associate Professor Yang Hong of the HKU Business School release the world's first report in the KYA field

Agents accelerate into the real economy as a trust deficit erodes the AI dividend

The report summarizes the evolution of agents into four stages: capability agents, execution agents, transaction agents, and economic agents. As of now, the industry as a whole remains in a stage of accelerated progression from capability agents to execution agents, with some frontier scenarios beginning to explore transaction agents.

Moving from executing tasks to cross-entity transactions means that agents need to obtain trust beyond an enterprise's internal processes. Procurement, payment, or professional-services agents can take actions on behalf of users within the scope of their authorization, but counterparties still need to confirm whom they represent, what authorization they have obtained, whether they can reliably perform, and who bears liability when losses occur.

Industry research shows that enterprises' willingness to deploy agents is rising rapidly. In its 2026 "Hype Cycle for Agentic AI," Gartner noted that 17% of surveyed organizations have already deployed AI agents, and more than 60% expect to deploy them within the next two years. Gartner also predicts that by 2028, Fortune Global 500 companies will use an average of more than 150,000 agents each.

The potential scale of agent commercial activity has also drawn attention. McKinsey estimates that by 2030, agent commerce could bring coordination revenue opportunities of $3 trillion to $5 trillion to the global retail market; this estimate currently covers only retail goods and does not include services or B2B markets.

In stark contrast to the rapid improvement in capabilities, the trust system supporting the operation of agents remains almost nonexistent. The report characterizes this misalignment between capability and institutions as the "agent trust vacuum": over the past two years, security incidents such as agents exceeding their authority, prompt injection, tool poisoning, and failures of liability in multi-agent collaboration have emerged intensively and have directly translated into measurable economic costs.

A survey of the global financial industry by the Cloud Security Alliance and Anjuna shows that 62% of surveyed institutions have deployed AI agents. Meanwhile, Accenture research found that 87% of surveyed financial institution respondents believe that "trust" will become the biggest obstacle to the large-scale application of agent payments.

In his presentation, Yang Hong noted: "Agents are entering the real economy, and their diffusion speed is far faster than the adoption speed during the internet development period. However, the trust deficit is eroding the AI dividend."

76% of surveyed enterprises have agent trust concerns, and trust friction becomes a hidden commercial cost

The industry questionnaire shows that agent trust friction has become normalized: 76% of surveyed practitioners encountered concerns related to agent trust, security, accountability, or compliance frequently or occasionally over the past year, with 23% encountering them frequently and 53% occasionally, while only 2% said they had never encountered them.

Trust concerns have also produced actual operational impacts: 41% of respondents mentioned increased compliance and audit pressure, 40% mentioned delays in project launch or expansion, 38% mentioned increased manual confirmation and reduced efficiency, and 23% said they had encountered refusal by customers or partners to use the technology.

Among the main obstacles to advancing commercial applications of agents, "low trust from customers or partners" (53%) and "unclear attribution of liability after errors" (42%) ranked as the top two, ahead of difficulties in system integration (35%) and difficulty in measuring return on investment (34%).

Main findings of the University of Hong Kong's "KYA: Research Report on the Agent Credit System"

Consensus on the demand side is equally strong. 88% of respondents believe that the industry needs unified standards for agent trustworthiness assessment or certification, and 60% are willing to proactively adopt related services or first pilot them in high-risk, critical scenarios.

The research team believes that the bottleneck in the development of the agent economy is shifting from purely technical capability to whether the market can judge and manage the action risks of agents. Without trustworthy identity, clear authorization, verifiable performance records, and liability arrangements, enterprises and consumers will find it difficult to confidently expand delegation, and agents will find it difficult to move from internal task executors to participants in cross-entity transactions. KYA thus becomes the core hub connecting agent technology with real economic activity.

"The main constraint on the agent economy is shifting from technical capability to the trust system," Wang Yang emphasized. "Once agents enter payments, the trust issue directly becomes a commercial issue. The goal of KYA is to safely unleash the scale of the agent economy."

"Trust capital" will affect the commercial value and competitive advantage of agents

Agents with similar technical capabilities may have different commercial value. The report argues that trusted identity, clear authorization, reliable performance records, and liability guarantees affect whether enterprises are willing to delegate and how much authority they are willing to grant.

The report accordingly proposes the concept of "agent trust capital": intangible capital that economic entities form through investment in trusted governance and through agents' continuous, verifiable performance activities, which can reduce transaction costs, expand the scope of economic activity, and create future returns.

Ant Digital Technologies' Agent Super Factory is one of the typical cases included in the report. Its Agentar platform covers agent development, deployment, operation, and governance. Relying on its long-term layout in blockchain, digital identity, privacy computing, and security risk control, Ant Digital Technologies' Agent Super Factory has incorporated identity verification, authorization management, and behavior records into its business processes. The report argues that agent platforms have the opportunity to embed trusted capabilities throughout the entire lifecycle, providing a foundation for the continuous accumulation of trust capital.

It is reported that during Hong Kong Fintech Week, Ant Digital Technologies released a KYA product upgrade, providing identity infrastructure for all products under its Agent Super Factory model.

During the 2026 Hong Kong Fintech Week x StartmeupHK Festival, Ant Digital Technologies showcased the KYA product upgrade

The report includes nine representative enterprises that are building agent trust, covering fields such as payments, finance, education, energy, and healthcare. The report forecasts that the industry will see four types of strategic opportunities: production agents—shifting from selling software tools to providing digital labor; organizational agents—shifting from managing software processes to organizing and configuring agent productivity; trust agents—evolving from reducing single-transaction risk to providing agent credit services and cultivating trust capital; and connecting agents—moving beyond serving enterprise digital transformation to further connecting cross-platform agent services and economic activities.

"Economic competition in the AI era will not only be competition among models, computing power, and algorithms, but will also increasingly manifest as competition in the trustworthiness of production capacity, organizational efficiency, and the scope of economic collaboration," Wang Yang said.

Leiphone Leiphone