Financial institutions are under siege from a wave of increasingly sophisticated fraud — much of which is powered by AI. Despite added layers of authentication like one-time passcodes, biometric checks and behavioral analytics, bad actors are adapting faster than ever. Even OpenAI CEO, Sam Altman, has warned that AI has defeated fraud detection tools like voiceprint, making it harder to distinguish real from fake in contact center interactions.

The Financial Services Information Sharing and Analysis Center (FS-ISAC) predicts that loss from deepfake and other AI-generated frauds will reach $40 billion in the US by 2027.

AI has not only raised the bar for scam sophistication. It has essentially industrialized fraud at a newly massive scale.

This accelerating game of cat and mouse has left many financial institutions unsure of where to invest their defenses. In this article, we break down the newest forms of AI-fueled fraud that are on the rise and share some of the most effective, real-world tactics for stopping them.

Types of Next-Gen Fraud

The ability to detect and block fraud is no longer just a compliance exercise. It’s part of a broader operational resilience strategy that protects both the institution and its customers, helping maintain trust even when disruptions hit. That’s why it’s mission-critical to understand the ways in which fraud is evolving.

Deepfake Account Takeover Attempts

Fraudsters are using AI-generated voice clones to impersonate customers with alarming accuracy. An international group of researchers found that simply re-recording deepfake audio with natural acoustics in the background allows it to bypass detection models at a higher-than-expected rate. Basically, this means how we’d been detecting deepfakes doesn’t cut it; the numbers back it up. Voice fraud has grown by 1,740% from 2022-2023 in North America.

These deepfake voices can bypass traditional knowledge-based authentication, voiceprint and even fool human agents. This allows fraudsters to gain access to sensitive financial accounts and execute unauthorized transactions.

Synthetic Identity Farming Via the Voice Channel

Fraud rings are increasingly using synthetic identities — blended profiles built from real and fake data — to open and age accounts, often slipping past initial fraud checks by behaving like legitimate users. These accounts may sit dormant for months before being activated for large-scale fraud, such as credit abuse, insurance scams or money laundering.

Now, with generative AI tools, bad actors are creating synthetic voices to match these fake profiles. They call into contact centers to “build trust” with agents, change account details or slowly gain permissions. In this practice, they present themselves as customers with plausible backstories and calm, believable voices.

Executive Voice Impersonation

In high-value fraud scenarios, attackers use AI to mimic the voices of C-level executives, such as CFOs or VPs of finance, using deep learning models trained on speeches, webinars or earnings calls. In Hong Kong, a finance worker was scammed into paying $25 million after mistaking a fraudster for his Chief Financial Officer in a video conference call. The scam plays on authority and speed, often bypassing internal verification due to the voice’s perceived legitimacy.

Anti-Fraud Best Practices

Layered Authentication Methods

There are several different ways contact centers have worked to protect account holders. Many of them can be categorized by:

  1. Something an account holder knows, such as apassword or security question
  2. Something an account holder has, such as a one-time passcode or mobile device
  3. Something an account holder is, which includes biological markers

Fraudsters may get past one layer, but each independent layer makes it exponentially more difficult and expensive for attackers to break simultaneously. Think of it as the difference between a locked door versus a locked door with an alarm, surveillance and a guard dog. The more friction that is created for bad actors, the safer the account holder transactions become. This layered defense strategy is also known as “defense in depth”.

Traditionally, contact centers have focused on the first two layers: knowledge-based authentication and one-time passcodes sent via SMS or app. But these methods are increasingly becoming more vulnerable. Data breaches can expose personal information, and SMS codes can be intercepted.

Last is the third layer: something you are. Technologies like voiceprint have existed for some time, and they offer a way to authenticate callers through voice.

However, voiceprint relies on enrollment. And with the fast development of synthetic voice, it’s greatly diminished efficacy.

This is where voice biometrics comes in — and where the future of fraud prevention is heading.

Voice Biometrics

Voice biometrics analyzes the unique physical characteristics in a person’s voice to identify caller-voice profile mismatch without requiring any enrollment from the account holder. That means no Personally Identifiable Information (PII) or Non-Public Information (NPI) needs to be collected or kept on file. This can reduce compliance risks and exposure in the event of a breach. Unlike knowledge-based methods, which can be guessed or stolen, voice biometrics analyzes a person’s unique biological characteristics. This creates a high-fidelity way to authenticate callers beyond just a voiceprint. And because the human voice doesn’t rely on device possession (like a smartphone), there’s no risk of SIM swapping or passcode interception.

When evaluating voice biometric solutions, look for those that:

  • Work passively and in real time, without disrupting the call experience
  • Require no storage of PII or NPI
  • Integrate seamlessly into your existing contact center systems

Voice biometrics offers a high-security, low-friction way to protect your customers, closing the gaps left by traditional methods and making it harder for fraudsters to break through.

Fraudster Watch Lists

Some security solutions today include a Watch List, which is essentially a more adaptive alternative to static fraudster databases. Unlike traditional databases that are prone to false positives, a Watch List evolves in real time, based on voice bio-signals. It doesn’t flag someone based on a single incident but rather tracks suspicious activity and behavior to assess risk more accurately.

Real-Time Fraud Scoring

Fraud scoring models now play a central role in security, assessing risk dynamically throughout the call. Outdated systems evaluate a range of inputs: behavioral cues like hesitations or shifts in tone, technical metadata such as call origin or duration anomalies, and historical indicators tied to known fraudulent activity.

Rather than relying on fixed rules, modern systems use machine learning to adapt in real time. They assign a fluid risk score as the conversation unfolds. Calls that cross a certain risk threshold can be automatically escalated into more rigorous verification flows. This allows businesses to intervene earlier, often before a breach occurs, while reducing friction for legitimate callers.

Looking Ahead to Mitigate Risks

As AI continues to reshape the threat landscape, fraud prevention can no longer rely on reactive tools or outdated verification tactics. It demands real-time, adaptive systems built to outpace machine-driven attacks.

Voice is quickly becoming the new battleground. And without layered defenses, every call becomes a potential risk.

For financial services institutions, trust is the cornerstone of customer relationships. In an environment where a mix of digital and voice interactions is the norm, integrating fraud prevention into the customer experience (CX) strategy is no longer optional; it’s essential.

A CX platform that incorporates anti-fraud capabilities ubiquitously across the ecosystem ensures that security and convenience work hand in hand. By proactively detecting and preventing fraud within every interaction, financial institutions can safeguard assets, protect reputations and reinforce regulatory compliance while delivering the seamless, trustworthy experiences customers demand.

Ready to evolve your customer experience with a AI-powered technologies? Read our “2026 Buyer’s Guide for AI and CX” to learn what defines future-ready CX platforms and see how leaders are uniting data, systems and people to deliver personalized experiences at scale. 

Jack Caven, CEO and Co-Founder of VoxEQ, also contributed to this article.

VoxEQ is a voice intelligence solution that recognizes key demographics of each caller in seconds. By unlocking the unique bio-signals found in the human voice, VoxEQ helps contact centers better detect fraudulent activity, route calls more effectively, and elevate customer conversations from the first hello.