How AI Agents Are Changing Australian Banking

Australian banks have used artificial intelligence for years, but AI agents mark a bigger shift. Instead of only analysing data or answering one question, these systems can follow a goal, gather information, use approved tools and complete several steps within a workflow.

For large banks, that could mean investigating suspicious payments, preparing case summaries, helping frontline teams retrieve policy information or guiding customers through routine service requests. The opportunity is significant, but banking is not an industry where autonomy can be added without strong controls.

From our perspective, the question is no longer whether AI agents will enter Australian banking. It is how banks can scale them without weakening customer trust, operational resilience or accountability.

Fraud Detection Is Becoming More Proactive

Fraud prevention is one of the clearest uses for AI agents because threats change too quickly for static rules alone. Australians reported $2.18 billion in scam losses during 2025, with investment scams, payment redirection and phishing among the largest categories. The ACCC has also warned that AI is increasing the sophistication and scale of scam activity.

Commonwealth Bank offers a useful local example. In April 2026, it announced an agentic AI system that monitors transaction and payment data for emerging fraud patterns. The system assesses suspicious activity and proposes new detection rules, but those rules still require approval from the bank’s fraud analytics team.

That human-in-the-loop approach matters. AI agents can examine large volumes of activity quickly, while experienced staff remain responsible for decisions that may block payments or affect customers. For enterprise banks, controlled autonomy is often safer than full automation.

Customer Service Is Moving Beyond Chatbots

Traditional banking chatbots are usually limited to scripted questions. AI agents can potentially manage longer service journeys. An agent could identify a request, check relevant policies, collect missing information, update an internal case and prepare the next action for an employee.

CommBank reported that its Compass AI tool has answered more than 500,000 questions since July 2024 and helps business-banking staff retrieve information three times faster than traditional methods.

Speed is only useful when the response is accurate. AI agents should not make unsupported promises, expose another customer’s information or provide guidance outside approved boundaries. Banks also need clear escalation points so customers can reach a person when a matter is sensitive, disputed or unusual.

For large enterprises, the goal should not be to remove people from every customer interaction. It should be to give employees better information, reduce avoidable delays and make routine service more consistent.

Back-Office Work May Deliver the Biggest Gains

Some of the most valuable changes may happen away from customer-facing channels. Banking teams manage repetitive work across onboarding, complaints, lending operations, financial crime investigations and regulatory reporting.

AI agents can assemble documents, compare information across systems, identify missing fields and draft internal summaries. An agent supporting a fraud investigator could retrieve transactions, organise alerts and prepare a timeline. The employee can then focus on judgement rather than administration.

This is especially relevant to large banks with complex technology environments. The value comes from connecting workflows, but those connections also create risk. Every system permission, data source and automated action needs to be deliberately limited.

An AI agent that can read a customer file presents one level of risk. An agent that can alter the file, contact the customer and restrict an account presents another. Enterprise governance must recognise that difference.

APRA Is Watching Governance Closely

Australian regulators are making it clear that existing responsibilities still apply when banks adopt advanced AI. In April 2026, APRA said AI was moving into operational and customer-facing applications faster than governance, assurance and resilience practices were maturing.

APRA also raised concerns about board capability, dependence on major technology providers and limited visibility into AI embedded within broader software platforms.

For banking leaders, AI agents cannot sit only within an innovation or technology team. Risk, legal, privacy, cyber security, procurement, data and business owners need shared visibility.

Boards also need enough understanding to challenge management on where AI agents operate, what information they can access, which decisions they influence and how failures would be contained.

ASIC has separately urged financial entities to strengthen cyber resilience as AI accelerates vulnerability discovery and attack activity. Its guidance highlights access reviews, third-party risk controls, incident response planning and protection of critical systems.

Privacy and Automated Decisions Need Attention

AI agents may handle identity information, transaction histories, complaints, credit data and behavioural signals. That makes privacy governance central to deployment.

From 10 December 2026, organisations covered by the Australian Privacy Principles will face new transparency obligations where personal information is used in automated decisions that could significantly affect a person’s rights or interests.

Relevant privacy policies will need to explain the types of personal information used and the kinds of automated decisions being made.

Large banks should prepare before that deadline. Customers need to know when automation is influencing an important outcome, how they can question it and where human review is available.

What Responsible Enterprise Adoption Looks Like

Banks scaling AI agents need a central register of use cases, named owners and clear risk classifications. Each agent should have defined permissions, approved data sources, testing requirements and a reliable shutdown process.

Monitoring should cover accuracy, bias, privacy leakage, security, unexpected tool use and customer impact. Logs must show what the agent accessed, recommended and completed.

High-impact actions should still require human approval, particularly where money movement, credit, account restrictions or customer rights are involved.

AI agents are changing Australian banking because they can move beyond advice and participate in real work. Their value will not be measured only by tasks completed or minutes saved. It will be measured by whether banks improve service and security while keeping responsibility firmly with the organisation.