← Back to News
Regulation

Hong Kong sandbox picks 36 agentic AI pilots, including agent ID

Four Hong Kong financial regulators named 36 agentic AI use cases for the first GenA.I. Sandbox++ cohort on August 27, 2026. One, from HKT Payment and Red Date Technology, tests how to register and verify the software agents that initiate payments.

Four Hong Kong financial regulators on August 27, 2026, named the first cohort of the GenA.I. Sandbox++, selecting 36 use cases from nearly 100 proposals. Several involve payments, and one takes on a question the industry has yet to settle: how to identify the software agents that initiate a transaction.

The Hong Kong Monetary Authority (HKMA), the Securities and Futures Commission (SFC), the Insurance Authority (IA), and the Mandatory Provident Fund Schemes Authority (MPFA) run the program with Cyberport, the territory’s government-owned technology park. The 36 use cases involve 30 financial institutions and 27 technology partners. Technical trials are due to start before the end of 2026 on a platform managed by Cyberport’s Artificial Intelligence Supercomputing Centre.

Sandbox++ opens the program to insurers and fund managers

The cohort is organized around three themes: risk management, anti-fraud measures, and customer experience. The original GenA.I. Sandbox opened in 2024 for banks only and focused on content generation. The HKMA announced a second cohort in October 2025. The Sandbox++ extension, launched in March 2026, brought in asset management, insurance, and mandatory provident funds, and shifted the trials toward agentic AI: systems that carry out tasks and make decisions without a human sign-off at each step.

36
use cases selected for the first cohort
HKMA, SFC, IA, MPFA, via SCMP
≈100
proposals received
SCMP
30
participating financial institutions
TechNode Global
27
technology partners
TechNode Global

The trials cover customer onboarding, payments, insurance claims, and customer interactions. Named participants include Ant Bank (Hong Kong), WeChat Pay Hong Kong, and HSBC Life (International). The cohort carries over the “A.I. vs. A.I.” approach of earlier rounds, in which one AI system checks the decisions of another, but now applies it to end-to-end processes rather than to single tasks.

Payment interface on a screen
The trials cover complete processes, from customer onboarding to payment.

Agentic ID ties each agent to a customer and a mandate

The project closest to payments pairs HKT Payment, the payments arm of telecom operator HKT, with Red Date Technology. The two companies are building Agentic ID, a framework based on decentralized identifiers and verifiable credentials that also uses zero-knowledge proofs. It is designed to register, then verify, the software agents that initiate a payment, a wallet top-up, a peer-to-peer transfer, or an interbank transfer.

Know-your-customer (KYC) checks establish that customers are who they say they are. They don’t answer the three questions an autonomous agent raises: whether the agent is linked to an identified customer, whether that customer authorized it, and whether the authorization is still valid at the moment the transaction goes out. Agentic ID is meant to support three controls:

  • a bounded mandate that sets approved merchants, accessible accounts, a per-transaction limit, and restrictions on location or dates
  • revoking an agent’s authority without closing the account of the customer it acts for
  • an audit trail that holds up in a dispute, showing which agent acted, for which principal, and under what instructions
🔑
What “know your agent” adds to “know your customer”
A traditional identity check applies to a person and is rarely repeated. A software agent adds a second layer, the mandate, which is created, narrowed, and revoked over the life of the customer relationship. The two layers answer different questions. An institution that manages only the first cannot tell, at the moment of payment, whether the agent placing the order is still acting within the limits the customer set.

Admission to the sandbox is not approval to launch

A regulatory sandbox lets supervised firms test a technology on live cases, within a limited scope and under the regulator’s watch. Getting in does not amount to approval to go live. Each system remains subject to existing licensing regimes and governance requirements, and any commercial rollout needs a separate decision from the relevant regulator.

Two questions remain open after the announcement. The first is whether an agent identity issued in Hong Kong will be recognized outside the territory. The second is how liability is split among the customer, the account-holding institution, and the agent’s developer when a formally authorized transaction turns out to be one the customer never wanted.

2024
GenA.I. Sandbox
The program opens to banks, focused on content generation.
October 15, 2025
Second cohort
The HKMA announces a new round of banking projects.
March 2026
Sandbox++
The scope widens to asset management, insurance, and provident funds, with a shift toward agentic AI.
August 27, 2026
First cohort
Four regulators select 36 use cases from nearly 100 proposals.
Late 2026
Technical trials
Trials begin at Cyberport’s Artificial Intelligence Supercomputing Centre.
ℹ️
How this story was sourced
Details of the individual projects come from consistent reports by the South China Morning Post, TechNode Global, PYMNTS, and Caproasia. The benefits attributed to the projects are those claimed by their sponsors, before any trial has run.

Agent identity is on the agenda of several other payments-industry efforts, from the mandate protocols promoted by the card networks to the agentic commerce alliances formed in 2026. Hong Kong’s approach stands out for its starting point: it puts the question inside a supervisory program rather than in a technical specification published by a private company. The trial results will show whether agent identity becomes a prudential requirement or remains a voluntary market practice.

Provenance

Published August 27, 2026

5 sources, 5 distinct domains

↗ South China Morning Post, “Hong Kong expands AI sandbox with projects that include banks, securities and insurance” · scmp.com↗ TechNode Global, “Hong Kong selects 36 agentic AI financial pilots,” August 27, 2026 · technode.global↗ PYMNTS, “Hong Kong Tests Know Your Agent Controls in Payment Workflows” · pymnts.com↗ Caproasia, “Hong Kong Announced First Cohort of GenA.I. Sandbox++ with 36 Use Cases in 3 Themes” · caproasia.com↗ Hong Kong Monetary Authority, “HKMA announces second cohort of GenA.I. Sandbox to advance responsible A.I. innovation,” October 15, 2025 · hkma.gov.hk
← All news