Over the past few years, the nature of financial crime has evolved more rapidly than most institutions can respond. Fraud is no longer just about stolen credentials or unauthorised card use. It’s about deception, which is often subtle, digital and increasingly difficult to detect with tools that may have been effective just a few years ago.
Criminals aren’t breaking in. They’re being invited in by the customer.
And that’s the big change. Fraud today is more human, more scalable and far more challenging to contain. As generative AI lowers the barrier to entry, the number of attacks and the speed at which they evolve are multiplying. Traditional fraud defences weren’t built for this.
At Feedzai, we monitor fraud trends across hundreds of institutions globally. What we’re seeing isn’t just a new tactic or spike. It’s a systemic shift in how fraud works and how it must now be prevented.
From stolen cards to stolen trust
Not long ago, the biggest threat was third-party fraud: someone stole a card, made purchases and left a digital trail. Banks can detect fraud using tools such as device fingerprinting, geolocation or behavioural analytics.
Now, the challenge is more complex. Fraudsters are bypassing these systems by directly manipulating customers. They persuade victims to send money, approve payments or share access, often while believing they’re protecting themselves.
This type of fraud, a scam, is challenging to detect because the activity often appears legitimate. It’s the customer’s device. It’s their credentials. It is, in fact, their transaction. In this scenario, the controls do not fail. They’re sidestepped.
Many customers validate the transaction themselves, thinking they’re speaking to their bank or someone they trust. You end up with fraud that’s harder to detect and harder to reverse and that obliterates trust.
Generative AI changes everything
We’ve all seen our feeds flooded with AI-generated videos that are becoming increasingly difficult to distinguish from reality. The technology is impressive, even thrilling. But for fraud teams, it’s a nightmare.
GenAI has lowered the cost of committing fraud while increasing its scale and sophistication. Common red flags, such as bad grammar, pixelated images and awkward phrasing, have all but disappeared. A polished phishing site can be spun up in minutes. A synthetic voice can convincingly mimic a loved one, a CEO or a bank employee.
And fraudsters no longer need technical skills; they just need prompts.
We’re seeing higher returns for criminals and significantly more volume. That’s why institutions need to adopt equally adaptive defences. Advanced fraud technology that learns quickly, detects early and distinguishes between benign and high-risk behaviour in real time.
Three critical priorities for AI fraud
First, treat fraud as a customer experience challenge
Security can’t be a background process. It must be designed into the experience. At Feedzai, we believe fraud prevention should be interactive, informative and empowering. That means:
- Give customers real-time control over their risk settings
- Provide contextual warnings, not generic blocks
- Help customers understand and avoid threats
- Build trust so that fraud protection isn’t just about stopping crime; it’s a way to deepen your relationships.
Second, AI fraud require AI defences
Feedzai’s RiskOps platform utilises hundreds of machine learning models deployed across more than 190 countries, detecting suspicious behaviour, identifying emerging fraud rings and enabling instant decision-making.
But it can’t just be about speed. It has to be about nuance, too. Large language models (LLMs) and agentic AI can help human fraud agents have more effective conversations with victims by understanding that a scam is an emotionally complex experience and adapting support accordingly.
Third, redefine remediation
Not all fraud is the same. Some cases can and should be automated. Others require empathy, skill and deep domain expertise.
Feedzai is pioneering a differentiated treatment strategy that routes cases based on three early indicators:
- Automated self-service for straightforward, low-risk issues
- Fraud expert intervention for emotionally complex or high-stakes scams
- Criminal investigation workflows for mule accounts
and synthetic identities
This three-tier approach ensures that victims feel heard, criminals get caught and resources are used effectively.
The future is adaptive, automated and human
The next phase of fraud prevention isn’t just about automation. It’s about smart automation paired with intelligent human intervention. Some cases require speed, while others require empathy and understanding. The key is knowing which is which and routing accordingly.
Banks that embrace this shift will deepen customer relationships and differentiate with trust. Those who delay may find that fraud is no longer just a cost. It’s a brand issue.
