The rise of the ‘digital colleague’?
How banks in Singapore put agentic AI to work
As the Monetary Authority of Singapore (MAS) tightens its regulatory oversight on artificial intelligence in the financial sector, banks are entering a new phase of AI adoption – one where digital systems are expected not just to answer questions, but to carry out increasingly complex tasks.
Rather than using AI just to generate content or respond to prompts, banks such as DBS, OCBC and UOB are redesigning workflows across areas such as wealth advisory, client onboarding, know-your-customer (KYC), compliance and operations around AI “agents”.
These agents can reason, plan and execute complex, multistep tasks with limited human intervention.
Decoding the differences
This shift from generative to agentic AI marks what many see as the next frontier of banking.
Generative AI: Capabilities
Generate outputs in response to prompts. Typically requires humans to decide and execute subsequent actions
Agentic AI: Capabilities
Pursue specified objectives by planning intermediate steps, selecting tools, and initiating actions without continuous human direction
Generative AI: Primary function
Create original content – such as text, images, video, audio or software code – in response to a user’s prompt or request
Agentic AI: Primary function
Designed to autonomously make bounded decisions and actions, with the ability to pursue complex goals under limited supervision
Generative AI:
Relies on machine learning models called deep learning models trained on large datasets
Agentic AI:
Combines the flexibility and natural-language reasoning of large language models with the reliability, determinism and guardrails of traditional software engineering
Generative AI: Key use
- Content creation
- Data analysis
- Personalisation
Agentic AI: Key use
- Decision making
- Problem solving
- Workforce automation
- Planning
Source: MAS, IBM
While banks differ in how they view these systems – from sophisticated assistants to “digital colleagues” – the technology could fundamentally reshape how work is organised, with AI agents taking on a growing share of tasks alongside employees.
Yet, risks abound. On Jul 22, OpenAI disclosed that its autonomous agents “went rogue” and independently hacked into another company’s systems during testing, underscoring the cybersecurity challenges that come with the technology.
The Business Times examines how banks are moving from generative to agentic AI, where these technologies are being deployed, and how institutions are aligning with MAS’ evolving AI governance frameworks.
Banks and their AI plans
The interactive overview covers Citi, DBS, Maybank, OCBC, Standard Chartered and UOB.
Interactive bank card carousel
Drag or flick · Tap any visible card · Arrow keys rotate · Enter opens
Jobs, jobs, jobs
The shift is also reigniting concerns over the future of banking jobs. While fears of job losses have accompanied the rise of agentic AI, banks generally frame the technology as augmenting, rather than replacing, jobs.
DBS chief executive Tan Su Shan has said AI will eliminate some aspects of jobs but allow employees to take on higher-value work, describing the bank's philosophy as protecting "workers, not jobs".
OCBC similarly expects roles to “naturally evolve and change” as new technologies reshape the workplace.
But workforce transformation is just part of the story. As banks give increasingly autonomous systems a bigger role in workflows, the challenge lies in ensuring they operate safely and transparently, so they become trusted digital colleagues.
AI governance
Against this backdrop, MAS on Jul 3 unveiled the Safeguards for Agentic Finance at Runtime (SAFR), an industry white paper providing a set of governance checkpoints that verify and record an AI agent’s proposed actions before it executes its tasks.
Agent identity
Is the agent a recognised, registered agent?
Controls repository
Which controls should the proposed action be checked against?
Disposition engine
How should this action be handled?
Audit log
Captures the record, decision basis, and outcome for every action regardless of result.
Source: MAS
Observers have described this as a "genuine turning point" in moving the financial industry from broad principles to operational safeguards.
"AI agents are no longer theoretical for financial institutions,” said Bryan Keasberry, Apac head of market development at compliance software provider Fenergo.
“Firms are already exploring how they can support client onboarding, compliance reviews, advisory workflows and operations, and the focus has shifted to how that can be done safely once AI starts interacting with live systems, data and decision-making processes.”
Chris Robinson, group chief technology Officer at IQ-EQ, agreed, saying that the conversation has evolved from whether institutions should use AI to how to deploy increasingly autonomous AI systems “safely, transparently and at scale”.
Timeline of some key initiatives on AI by MAS:
-
MAS consultation on AI risk management closed
Oversight of AI risk management, policies and procedures, key AI life cycle controls
-
MindForge AI Risk Management Toolkit launch
Developed collaboratively with 24 leading financial institutions to manage AI risks
-
Generative AI Guardrails in Banking Handbook published by MAS and ABS
Framework for implementing Gen AI safely across financial institutions
-
Harness AI in the Fight Against Financial Crime, in collaboration with the banking industry
Use AI and machine learning to enhance scam detection capabilities
-
Future of Finance Institute (FFI) announced
National innovation centre driving the financial sector's transition from AI and tokenisation
-
SAFR Framework (BuildFin.ai) white paper published
Framework for the governance of AI agents in financial services
Source: MAS
Challenges with agentic AI
Robinson said the industry needs to distinguish more clearly between different categories of AI.
“The risks associated with an AI tool that extracts information from documents are very different from those of an AI agent providing recommendations or supporting client-facing decisions,” he noted.
Keasberry meanwhile highlights three main challenges companies face in implementing agentic AI:
Underlying infrastructure
Many institutions still operate with fragmented customer data, legacy systems and manual processes spread across teams and jurisdictions. This complexity could be exacerbated when AI agents are introduced.
Governance
AI adoption cannot sit with technology teams alone. Risk, compliance, operations and business functions all need to agree on where AI can be used, what level of human review is required at each decision point, and how outcomes are monitored over time. Getting that alignment across an institution is harder than building the technical capability itself.
Regulatory engagement
Supervisors across the region are at different stages of readiness to accept agentic decision-making in regulated activities. Institutions will need to demonstrate not only that controls like SAFR are in place, but that those controls are effective, explainable and subject to genuine human accountability. That requires early, open dialogue with regulators rather than a compliance-filing approach.
Beyond these challenges, Keasberry noted that many banks are still running AI pilots in controlled environments. The bigger challenge, he said, lies in embedding AI into regulated workflows where outputs affect customer risk ratings, due diligence outcomes or advisory recommendations.
“This is seen as a significant opportunity, though speed cannot come at the expense of governance.
Robinson agreed, noting that the winners in the next phase of AI would not necessarily be those deploying it the fastest, but those demonstrating strong AI governance, clear accountability and robust controls alongside innovation.
Whether banks ultimately view AI agents as sophisticated assistants or digital colleagues, their value will depend on earning the trust of both clients and regulators.