Continuous adaptive trust is the new fraud detection mandate

By Dan Holmes, VP of Product Planning and Strategy at Feedzai

The fraud prevention landscape is transforming at a rapid pace. New innovations like AI are helping bad actors to supercharge their attack schemes and outpace traditional defenses. According to the Global Anti-Scam Alliance’s 2025 Global State of Scams report, victims lost $442 billion to scams across the 42 countries reflected.

According to recent research from Gallup and Stop Scams Alliance, 12 percent of successful scams in 2025 involved AI. That figure—and the financial damage associated with it—is only likely to grow over time. As fraudsters target more victims in more sophisticated ways and on a larger scale than ever before, the financial world needs to prioritize continuous adaptive trust to stay safe.

Continuous adaptive trust treats identity as a probability rather than a verdict. Without it, fraud prevention risks becoming a reactive practice that draws on past trends, rather than a proactive safeguard that anticipates new threats. A correct password proves someone typed the right characters. It does not prove who typed them or their intent. Confidence is then built from the right available signals (the device and situation, the behavior, the credential outcome, and shared risk signals), re-evaluated continuously, and weighed against what the customer is actually trying to do. Friction appears only when risk outruns confidence.

How AI is transforming identity verification

Fast-evolving AI is the most significant driver of the shift to continuous adaptive trust. Able to act autonomously and replicate human behavior across text, voice, and video, AI agents, deepfakes, and other innovations have shattered the traditional identity verification paradigm, and the technology is only getting better.

Recent Anthropic models like Mythos and Fable have been rolled out slowly due to worries about the damage they could cause in the wrong hands. In the case of Mythos, those risks were inherent to the banking system. Able to readily spot hidden bugs and flaws, the model threatened to expose vulnerabilities in banking technology and cause upheaval in the digital systems that underpin our entire financial lives.

As a result, the future of banking requires an identity verification system that avoids overreliance on any one signal or score. Even biometrics, traditionally inextricable from a user’s identity, can be deepfaked through AI technology. By simply gaining account access with the help of AI, fraudsters can quickly exploit a trove of valuable and personal information.

The pitfalls of static identity signals

Static identity signals like passwords and credentials were once a highly effective way to stop or inhibit financial crime, but defenses can never stay complacent in the long term. Fraudsters hone their skills, methods, and technologies to exploit the weaknesses of every new defense. Protecting customers’ financial well-being is, and always will be, an adaptive practice for banks.

For example, bad actors have long employed phishing schemes to steal passwords, and they’ve disguised their intentions more skillfully with the help of AI. In response, organizations implemented multi-factor authentication (MFA) to offer strong resistance and introduce an added layer of complexity for fraudsters. Now, passkeys are a fast-growing way to keep accounts and sensitive information safe.

This is just one example of how banks and enterprises must respond to fraudsters’ encroachments, but it points to a larger truth. Relying on just one signal simply puts risk at an unacceptable level. It is easier than ever for criminals to bypass individual defenses. Adding layered defenses creates a compounding problem for them and makes it less likely that they’ll breach sensitive systems. By reducing the attack surface, financial institutions can improve the chances of stopping fraud or scams before money moves.

In practice, that might look like prioritizing passkeys over passwords, adding behavioral biometrics as a predictable layer to authentication, or examining behaviors both specific to the individual and representative of wider network trends.

Why continuous adaptive trust is the way forward

The end goal here is to build a more resilient architecture that continuously verifies identity and intent based on a holistic picture, not a single signal that can be stolen, deceived, or otherwise compromised. Continuous adaptive trust is exactly that, and its defining traits are ultimately in the name: “continuous” and “adaptive.”

It goes beyond simply identifying a user and dynamically discerns intent based on contextual or behavioral signals like the device in use. From there, it ensures that the correct permissions are in order. This is a constant process that analyzes behaviors in real time and sharpens signals based on demonstrated patterns like users’ typing cadences, devices and networks, and MFA history to build an ironclad barrier around sensitive financial information.

As threats emerge and iterate more quickly with the help of AI, continuous adaptive trust will be a de facto requirement for banks. According to IC3, internet crime losses ballooned by roughly 25 percent between 2024 and 2025, an increase that points to continued acceleration since AI became widely available in 2023.

Banks cannot afford to be caught flat-footed as bad actors use AI tools to evade traditional defenses and steal money from customers. Keeping funds and, most importantly, customers safe in today’s landscape means combining behavioral, situational, and shared risk signals with credential outcomes to make legitimate user behavior harder to fake and sensitive systems harder to access.

About the author

Dan Holmes is VP of Product Planning and Strategy at Feedzai. He is an experienced fraud prevention leader with a career working across multiple business sectors, including banks and fraud technology providers.

Article Topics

adaptive authentication  |  continuous verification  |  digital trust  |  Feedzai  |  fraud prevention