Hong Kong SFC's First AI Survey: 95% of Institutions to Accelerate Deployment as Regulators Focus on Personnel Accountability
nashnova research
The Hong Kong SFC surveyed 20 major financial institutions and found 95% plan to accelerate AI deployment in the next 12 months — but the regulator made one thing clear: no matter how smart the system, a human must answer for what it does.
How deeply has AI penetrated financial institutions?
The survey covered 20 large financial institutions and their 78 Hong Kong-licensed subsidiaries, conducted July–August 2026.
90% have already connected AI tools to internal systems; the same share is using frontier AI models for cyber defence or assessment.
This means → AI is no longer a pilot-stage experiment — it is already wired into day-to-day operations.
What is "agentic AI," and how many firms are using it?
Agentic AI — systems that can make decisions and execute tasks on their own, not just answer questions — was a key finding of the survey.
30% of firms already use it in live operations; another 25% are running pilots — more than half in total.
Use cases span data retrieval, content drafting, software development, operational workflows, and some trading-related activities.
In plain terms = these AI systems don't just look things up for you — they act on your behalf and make some decisions themselves. That is exactly what makes regulators nervous.
What is blocking large-scale rollout?
The survey identifies four bottlenecks: technical integration, moving pilots into production, uncertain cost-to-value, and data constraints plus talent shortages.
As AI applications multiply, firms also need to expand governance and monitoring to keep pace.
This reflects a status quo where AI can work in a single use case but struggles to scale across the full chain.
What is the SFC's regulatory stance?
Executive Director of Intermediaries Ye Zhiheng stated the approach clearly: "safety first, human-centred, collaboration-led."
He stressed that as AI autonomy rises, human primacy must not be undermined.
This means → the regulatory floor is not "don't use AI" — it is "for everything AI does, a specific person must be accountable."
What exactly will the regulator focus on?
The policy agenda targets four areas: investor protection, cyber security and operational resilience, human accountability and oversight, and industry knowledge-building.
The SFC specifically noted that the goal is not to require human sign-off on every AI output — controls should be proportionate to the AI's authority and potential impact.
In plain terms = low-risk tasks can run autonomously; high-risk decisions need a human in the loop — and someone identifiable when things go wrong.
What should licensed firms watch next?
The SFC will use the survey findings to develop formal regulatory guidelines.
The key question: can the rules keep pace with agentic AI's rapid iteration cycle? If guidelines arrive too slowly, firms have no compliance anchor to build against.
This reflects a deeper tension: the speed of AI evolution vs the speed of regulatory updates — and how Hong Kong resolves it will determine whether the city can move fast on AI finance without losing control.
市场有风险，内容仅供研究参考，不构成投资建议。