The Next Phase of AI in Financial Services

By Rohanshi Vaid

Artificial intelligence is no longer new to financial services. Banks use it to detect fraud, insurers to assess risk, and financial institutions increasingly rely on it to analyse information, support compliance and improve customer service.

What is changing is the role AI is expected to play. The next phase will move beyond tools that help employees work faster towards systems capable of analysing, recommending and, in some cases, acting with a greater degree of autonomy.

For financial institutions, this creates considerable opportunity. But it also raises a more difficult question. As AI takes on more responsibility, how much decision-making can genuinely be handed over to a machine?

From assistance to action

Until recently, many financial-sector AI applications remained largely supportive. A system might identify a potentially fraudulent transaction, summarise a document or provide an analyst with additional information, while a person remained responsible for the ultimate decision.

Agentic AI could change that relationship. AI agents are designed to carry out multi-step tasks with less continuous human direction. Applied to finance, such systems could eventually initiate transactions, manage parts of an investment portfolio, assess applications or complete operational processes on behalf of institutions and customers.

Singapore is already considering this next stage. In July 2026, the Monetary Authority of Singapore, together with financial institutions and fintech firms, published safeguards for the use of AI agents in finance. Its proposed framework includes controls around what an AI agent is permitted to do, real-time validation of its actions, auditability and mechanisms to stop or escalate activity when necessary.

The significance goes beyond efficiency. Once AI can act rather than simply advise, governance must increasingly operate while the system is running, rather than only when it is designed or reviewed.

The human in the loop is not disappearing

This is why greater AI autonomy is unlikely to mean the disappearance of human oversight in financial services. In fact, the opposite may be true.

Financial decisions can directly affect an individual’s access to credit, savings, insurance or investment products. Errors can also create wider operational and financial-stability risks. For these reasons, regulators are increasingly focusing on where humans must remain responsible and when they must be able to intervene.

Vietnam provides a particularly clear example. Its Decision 33/2026/QD-TTg, which takes effect on 15 August 2026, identifies high-risk AI systems across six sectors, including banking. The list covers AI systems that automatically execute high-value banking transactions without prior human review and systems that automatically make credit decisions. The Decision also states that AI systems must retain human supervision, control and intervention capabilities.

This does not necessarily mean that every AI-generated decision will require manual approval. Such a requirement could remove much of the efficiency AI provides. Instead, the challenge will be designing meaningful human oversight: deciding which decisions require approval, when AI should escalate a case, and who remains accountable when something goes wrong.

Financial services is becoming an early test for AI regulation

These developments also point towards a broader regulatory trend. Financial services is becoming one of the sectors where governments are moving most quickly from broad AI principles towards practical governance requirements.

Thailand’s Draft AI Act, released for consultation in July 2026, proposes a risk-based and sector-oriented framework under which relevant regulators would play an important role in governing higher-risk AI applications. Singapore, meanwhile, has largely developed its approach through its financial regulator. MAS has moved from broad principles on responsible AI towards practical tools for financial institutions, including its latest work on agentic AI. Vietnam has gone further by explicitly identifying particular banking applications as high-risk under its national AI framework. The approaches differ, but the direction is similar. Regulators are increasingly looking beyond whether AI is accurate or transparent and asking how it is governed when it affects consequential financial decisions.

These developments also highlight the policy challenge ahead. The task for policymakers is not simply to introduce more AI regulation, but to determine where existing financial-sector rules remain sufficient and where AI creates genuinely new risks. A risk-based approach will be important. AI used to summarise internal documents should not necessarily face the same requirements as a system making credit decisions or executing high-value transactions. Regulators will also need to provide greater clarity on human oversight, accountability, auditability and the use of third-party AI providers.

Singapore, Vietnam and Thailand already offer different examples of how this could work in practice. Rather than treating AI governance entirely separately from financial regulation, governments are increasingly combining economy-wide AI frameworks with sector-specific oversight.

What comes next

For financial institutions, the next stage of AI adoption will therefore be as much about governance as technology.

Firms will need a clear picture of where AI is being used, what authority individual systems have, which decisions can be automated and where human intervention remains necessary. Boards and senior management will also need to understand not only the risks of individual models, but how AI fits into existing frameworks for operational resilience, consumer protection, cybersecurity and accountability. AI will undoubtedly automate more of finance. But greater autonomy does not remove the need for human responsibility.

The institutions that navigate the next phase successfully will be those that can use AI to make financial services faster and more efficient while remaining clear about where the machine stops and human accountability begins.