Regulatory news and analysis across global markets

Hong Kong survey finds 13% of financial firms already run agentic AI, and 56% plan to

An HKIMR survey released by the HKMA finds GenAI use or plans at 90% of Hong Kong financial firms, with explainability the main barrier to agentic AI.

At a glance

Published

Regulators

Reading time

4 minute read

Sources for this story are listed at the end, with links to the original documents.

A night view of illuminated skyscrapers on Hong Kong Island, with the tallest tower in the centre, the angular lit outline of another tower on the left and dark hills behind.

The Hong Kong Institute for Monetary and Financial Research (HKIMR), the research arm of the Hong Kong Academy of Finance, on 17 November 2026 released a 75-page report on AI adoption in Hong Kong’s financial services industry, published through the Hong Kong Monetary Authority (HKMA). The survey behind it finds that 13% of financial institutions have already put agentic AI into some of their operations and 56% plan to. The agentic AI financial services survey also shows where firms want to use it first: compliance work such as Know Your Customer checks, anti-money laundering and transaction monitoring. In RegMemo’s analysis, that puts the first wave of autonomous AI inside the functions supervisors examine most closely, at a time when the HKMA is still consulting on how to supervise it.

Agentic AI in financial services follows near-universal GenAI use

The HKIMR commissioned the GenAI and Agentic AI in Financial Services Survey with KPMG Advisory (Hong Kong). Questionnaires went to banks, insurers, asset and wealth managers and technology and Regtech service providers between January and March 2026, and the HKIMR held 19 interviews between January and July 2026, 13 with financial institutions and six with Regtech providers.

Current or planned use of generative AI at the enterprise level reached 90%, up from 75% in the HKIMR’s 2025 survey. In 2025 respondents had projected 87% within three to five years, so adoption has already passed that forecast. Only 10% of respondents have not put GenAI to institutional use, citing resource and data constraints.

For agentic AI, systems that can interpret objectives and carry out multi-step workflows with limited human input, the split is 13% already using it to some extent, 56% planning to and 31% with no immediate plans but monitoring developments. Asked about investment priorities, 56% of financial institutions would prioritise GenAI while exploring agentic AI, 26% would take a balanced approach and 11% would put agentic AI first.

The first agentic use cases sit in compliance and back-office work

Financial institutions said they would deploy agentic AI first in internal-facing, process-heavy work: middle- and back-office operations, KYC and AML, and risk management. In compliance specifically, 86% of institutions have adopted or plan to adopt GenAI, with financial crime prevention and regulatory monitoring the leading current uses. More than 70% are interested in agentic AI for customer onboarding and due diligence, and for transaction monitoring and fraud detection, over the next three to five years. Some 73% plan to increase spending on GenAI for compliance.

Over half of institutions cited explainability and operational risk as major obstacles to agentic AI, and the GenAI challenges listed include hallucination, bias, data privacy, cybersecurity and data availability. Service providers added concerns about orchestration complexity, the difficulty of coordinating several AI agents on one task.

For a bank, an agent that screens onboarding files or clears transaction alerts makes decisions that end up in an AML audit trail. The report’s answer is human oversight scaled to the materiality, risk, autonomy, complexity and explainability of each use case, and it sets out three models: human-in-the-loop, human-on-the-loop and human-in-command. The last of these, where people set the limits of autonomy and the guardrails, is the model it links to agentic AI.

Hong Kong regulators are writing the supervisory rules in parallel

The report sits alongside a run of regulatory work this year. In October 2026 the HKMA announced its “ART” framework, covering accountability, resilience and trust, with updated supervisory guidance setting nine high-level principles for responsible AI adoption by banks. It also published a discussion paper on a proposed supervisory framework for agentic AI governance, which explores intent-anchored accountability, a registry of material AI agents and intervention mechanisms.

During Hong Kong FinTech Week 2026, the HKMA launched Project SAGE on agentic finance, and the Securities and Futures Commission (SFC) presented the results of its first Large Firm AI Readiness Survey, which will inform an SFC supervisory approach it describes as safety-first, human-anchored and collaboration-led. The GenA.I. Sandbox++ programme, announced in March 2026 by the HKMA, the SFC, the Insurance Authority and the Mandatory Provident Fund Schemes Authority with Cyberport, picked a first cohort in August 2026 focused on agentic AI, including the use of AI to oversee other AI applications.

The report’s five recommendations follow the same line: maintain regulatory clarity and proactive guidance, protect data privacy and cybersecurity, regularly reassess AI oversight and risk management, strengthen the resilience and capacity of computing systems, and build talent and cross-sector collaboration. Survey respondents asked for a layered approach, with high-level AI principles backed by guidance for specific use cases, and for clarity on accountability, explainability and third-party risk.

What to watch

The report sets no deadlines. The next steps sit with the regulators, and most have no published date:

  • The HKMA’s discussion paper on supervising agentic AI in banks, including a possible registry of material AI agents
  • The HKMA’s Cyber Resilience Testing Framework, still being developed, which will test how institutions respond to and recover from breaches
  • The SFC’s supervisory approach to AI, which its Large Firm AI Readiness Survey will inform
  • Results from the first GenA.I. Sandbox++ cohort on agentic AI, selected in August 2026

Primary sources

Related: HKMA asks banks whether AI agents need a central registry and a Know Your Agent standard


Photo (cropped): Mstyslav Chernov, CC BY-SA 3.0 via Wikimedia Commons

To Readers of RegMemo

Re The week in financial regulation

When Fridays

The weekly memo

One email at the end of each week: the rules, consultations and enforcement actions we covered, grouped by jurisdiction, with links to the regulators’ own documents.

Latest from the wire