At the 2026 Inclusion Bund Conference in Shanghai, two announcements pointed in the same direction: agents are being treated as accountable participants rather than anonymous scripts. Ant Group unveiled APASS, a trust infrastructure for agent commerce, and the Beijing Fintech Industry Alliance published China’s first group standard focused on the security of financial AI agent applications. Together they answer a question that did not exist a few years ago — when software acts on a person’s behalf, who is accountable, and for what exactly?
Key facts
| Fact | Detail |
|---|---|
| Question answered | China’s Agent Trust Infrastructure and Its First Financial AI Security Standard |
| Short answer | At the 2026 Inclusion Bund Conference, Ant Group launched APASS, a trust layer for agent commerce, while Beijing’s fintech alliance published China’s first security standard for financial AI agents. |
| Reading time | ~3 min |
| Last updated | 2026-09-12 |
APASS: learning who an agent is, and what it may do
APASS is built on the Know Your Agent idea. Rather than a one-time identity check, it maintains two trust chains, one for identity and one for behaviour, and packages them into identity registration, continuous verification, intent security and trusted evidence. In practice that means each party in a transaction can confirm which agent is acting, on whose behalf, with what authorization, and whether it is currently trustworthy.
Ant Group’s AI payment security lead Chen Shupeng framed agents as a new population in the commercial world: when they start calling services and joining transactions, the industry has to answer both who is trading and why. The reported scale suggests this is not a pilot — more than one million agents and over 50 institutions are said to be connected, and Gartner is cited forecasting that by 2028, 90% of B2B purchasing will be transacted through AI agents as intermediaries, moving more than 15 trillion dollars.
The financial standard: dual authorization
The more concrete rules come from the standard published on August 27, 2026. Its headline requirement is that third-party agents running on a phone may not use system permissions to automatically read or operate the graphical interface of a financial app without that institution’s authorization. Where an agent obtains data through a microphone, screenshots, screen recording or screen sharing, it has to respect the security policy of the party being called. Commentators summarize the rule as dual authorization: an agent operating a financial app needs consent from both the user and the institution.
The standard was led by the Beijing National Fintech Certification Centre with participation from Postal Savings Bank of China, ICBC, UnionPay, Bank of China, Bank of Communications and Huaxia Bank alongside Huawei, Ant Group, Volcano Engine and Tencent Cloud. It covers five areas: initialization and input, model inference and decision-making, identity verification and operation, data security and privacy protection, and risk control and compliance.
That rule is a response to a real incident. In December 2025 a phone shipping an AI assistant triggered login failures and blocked payments in multiple banking apps; that same month the assistant disabled its financial-app operation feature. The common industry answer since has been that operations must be permitted by the application owner — either through a provider-supplied interface or an MCP-style protocol that decides what is exposed.
Why the direction is familiar
Both announcements describe the model MOOGH already uses on the desktop: the agent proposes, the person approves, the action is scoped to what was authorized, and the result is recorded so it can be reviewed afterwards. Chinese regulators have been moving the same way, from the May 2026 joint implementation guidance on agent permissions and behaviour control to the July 2026 national guidance documents on agent interconnection and the mandatory standard plan for baseline agent application security.
Sources: Leiphone on APASS and Leiphone on the financial standard.
Download: MOOGH
Related reading: