Fraud is growing, and criminals now use generative AI too. Machine learning and AI agents give banks stronger defenses, with people accountable for the decisions that matter.
AI and machine learning help secure financial institutions in two ways. Machine learning models score transactions and customer behavior in real time, learning fraud patterns from large volumes of transaction data. AI agents then take on the work that follows an alert: they gather evidence across systems, validate information, and prepare cases for review. Together, these capabilities help banks respond faster, reduce false alarms, and protect customer trust.
Fraud grows as more of our lives, transactions, and ways of connecting move online. In 2025, U.S. consumers reported $15.9 billion in fraud losses to the Federal Trade Commission, up from more than $12 billion in 2024. About half came from investment scams, in which criminals persuade customers to send the money themselves. Those payments can look legitimate to controls built to catch stolen cards and credentials.
Executives at financial institutions across Latin America, Europe, and Asia-Pacific are moving beyond the question of whether AI belongs in their strategy. They want to know where it can create value first, how soon it can modernize operations, and how it can reduce complexity.
What is AI fraud detection in banking?
AI fraud detection applies problem-solving and pattern recognition to spot suspicious behavior across payments, accounts, and customer onboarding. Agentic AI is the newest layer. Unlike traditional AI tools that respond to questions or complete isolated tasks, AI agents coordinate activities across multiple systems. They gather information, validate data, trigger workflows, monitor exceptions, and support end-to-end business processes such as onboarding, servicing, and fraud operations.
Machine learning remains a critical fraud prevention tool
Machine learning adoption to combat fraud varies because institutions start from different points, shaped by their data, platforms, budgets, and skills.
Many financial institutions benefit from a trusted partner to help them adopt machine learning technology and enable real-time monitoring. That matters more as instant payments settle in seconds, leaving little time to stop a fraudulent transfer.
Machine learning models learn from known fraud patterns. New types of fraud, designed by bad actors but not yet seen, require a different approach. To combat this, some financial institutions are creating synthetic data on unique fraud patterns to train their machine learning systems. Synthetic data can augment a financial institution’s existing dataset of fraud behaviors to improve its overall fraud detection models.
Transaction data alone also misses many scams. When a criminal coaches a customer into sending money, the payment can look routine. Banks catch more of these cases by adding signals such as a new device, a first-time payee, or a receiving account that behaves like a money mule, one used to move stolen funds.
Criminals use generative AI too
Generative AI has also given fraudsters new tools. The U.S. Department of the Treasury’s March 2026 National Money Laundering Risk Assessment includes a special focus on AI in fraud and scams. According to the Federal Bureau of Investigation (FBI) information it cites, criminals use AI-generated text, images, audio, and video to commit fraud on a larger scale and make their schemes more believable. The assessment also notes warnings that criminals can use deepfake media to bypass customer identification checks at U.S. financial institutions, a risk the Financial Crimes Enforcement Network flagged in a November 2024 alert. A July 2026 FBI warning describes deepfake videos of a senior FBI leader that send people who already reported fraud to a spoofed complaint website. Together, these reports point to AI-enabled impersonation becoming a routine part of how financial fraud operates.
That makes detection harder. One-time identity checks and static rules are less reliable when a criminal can generate a convincing document, photo, or voice.
Agentic commerce is one of the most discussed fraud topics of 2026. AI agents are arriving on the customer side too, as payment providers build ways for consumers to let agents shop and pay for them. Fraud teams will need to tell a legitimate agent acting with the customer’s permission from a bot impersonating one, which is the goal of emerging “know your agent” practices.
How agentic AI changes fraud operations
Here’s the good news for fraud teams: an agent can handle much of the preparation work. Think of it as a digital coworker. It takes on information-gathering and process coordination so that fraud analysts, risk managers, and compliance specialists can focus on higher-value work.
Consider a fraud investigation. An agent collects the relevant documentation, validates identity information, summarizes account activity, identifies missing information, performs preliminary compliance checks, and prepares a recommendation package. The final decision still belongs to the bank.
This works best when intelligence connects to the platforms banks already run. Fraud signals appear across core banking, digital channels, and workflow systems. A system of intelligence links them, using AI, data, and automation to connect information, recommend actions, and coordinate work. Its immediate value comes from making existing systems more useful and responsive, so institutions can modernize without replacing the technology they already own.
“Technology is rarely the only challenge in transformation. The harder work is integrating innovation into everyday banking operations, with clear ownership, sound controls, and a realistic path to adoption.”
Jorge Andres Gomez, senior director, banking and financial services solutions, Unisys
Why security now sets the pace for AI
Unisys AI & Cloud Insights Report 2026 surveyed 1,000 senior IT and business decision-makers in the United States, Europe, and Asia-Pacific. It found that 49% of organizations suffered a cybersecurity breach in the prior 12 months, up from 17% a year earlier. Financial services was among the hardest hit, at 56%.
Security also shapes how much autonomy organizations give AI. Chris Arrasmith, executive vice president and chief operating officer at Unisys, puts it this way: “There are two sides of agentic AI: using it as an advisor and using it as an actor.” Starting as an advisor, with people making the decisions, builds trust before agents act on their own.
Practical steps for financial institutions
- Start with one measurable fraud use case, such as alert triage, onboarding identity checks, or investigation case preparation, and define what success looks like.
- Build on the platforms you already run by adding a system of intelligence layer instead of replacing core systems.
- Keep people accountable for consequential decisions, start AI agents as advisors, and expand their autonomy as trust and controls mature.
- Embed governance, security, and compliance in the design from the start, and test models against new tactics, including deepfakes and synthetic data.
- Fraud will keep changing as criminals adopt the same AI tools banks use. The banks that stay ahead will treat machine learning, AI agents, and human judgment as one connected defense, built on the platforms they already run.
How Unisys supports financial institutions
Fraud, onboarding, and servicing teams need intelligence that works with their current platforms and controls. Unisys brings together banking solutions, digital lending expertise, managed services, automation frameworks, and a growing focus on agentic AI. We help financial institutions identify where intelligence can improve fraud management, onboarding, lending, customer servicing, and risk operations, then integrate it into the environments they already operate.
Explore Unisys financial services solutions
Explore Unisys AI solutions