Generative AI in fraud detection: how to stay ahead of fraudsters
Fraudsters were early adopters of generative AI, and the institutions defending against them have been slower. That gap is the whole problem.
The question is no longer whether to use AI in fraud detection. It is which parts of the workflow to hand over, what governance has to exist first, and how to prove to an examiner that the output is sound.
We put that to three people working the problem from different angles:
Dana Lawrence, Sr. Director of Fintech Compliance at Pacific West Bank; Hailey Windham, podcast host and credit union advocate at Bank of Fraudology; and Matthew Hogan, Detective with the Connecticut State Police. Their answers are below.
What’s Holding Companies Back from Investing in AI in Risk and Compliance?
Many companies are hesitant to invest in AI for risk and compliance due to a variety of challenges:
Regulatory Concerns
Insights from Dana Lawrence, Sr. Director of Fintech Compliance, Pacific West Bank
Regulators have made it clear that existing laws still apply to AI in terms of risk and compliance. For instance, the Federal Deposit Insurance Corporation (FDIC) has shown openness to new technologies but expects risks to be managed appropriately. But, the challenge is that compliance teams must balance innovation with regulatory expectations. Institutions must ensure their AI in risk and compliance programs are transparent and explainable to pass regulatory scrutiny.
Job Preservation
Insights from Hailey Windham, Podcast Host & CU Advocate, Bank of Fraudology
Job preservation concerns around AI are misplaced. AI is a tool to enhance, not replace, fraud fighters. And if your current technologies don’t offer AI, that’s when you need to make that exit plan, and begin searching for another provider. If we’re still fighting AI-driven fraud with manual processes, it’s not going to work.
Evidence Admissibility and Accuracy
Insights from Matthew Hogan, Detective, Connecticut State Police
In law enforcement, AI adoption raises concerns about evidence admissibility and accuracy. Fraudsters operate without restrictions, while investigators must prove data integrity. Nevertheless, generative AI in fraud detection can be a game-changer for detecting large-scale fraud schemes, making adoption necessary despite its challenges.
The Growing Threat of AI-Driven Fraud and Its Challenges
The rise of AI-driven fraud presents an increasingly complex threat, raising several challenges for businesses to address:
The Slow Adoption of AI in Risk and Compliance
Insights from Matthew Hogan, Detective, Connecticut State Police
The adoption of AI in risk and compliance will be slow, as many companies, agencies, and law enforcement remain hesitant to fully embrace generative AI in fraud detection. This is often due to a lack of understanding of how to use it effectively. However, the rise of AI-driven fraud is inevitable, with everything from pig butchering scams to phishing and smishing being enhanced on a much larger scale.
Many still rely on KYC (Know Your Customer) processes to prevent fraud, but this approach is increasingly insufficient. While KYC remains essential, it doesn't address the complexities of modern fraud detection, as I've seen firsthand in cases involving subpoenas. It's clear that over-relying on KYC as the sole deterrent is no longer enough.
Challenges in Real-Time Detection and Response
Insights from Hailey Windham, Podcast Host & CU Advocate, Bank of Fraudology
As technology defenses improve, fraudsters quickly adapt, using generative AI tools that can bypass security checks and authentication measures in real-time. One of the most concerning developments is their ability to generate highly realistic synthetic identities with fake yet convincing data.
This complicates efforts, especially for teams operating in manual environments, as they struggle to keep up with these sophisticated threats' growing scalability and adaptability. However, the challenge is that these updates can’t always be implemented swiftly.
Rising Sophistication in AI-Driven Fraud
Insights from Dana Lawrence, Sr. Director of Fintech Compliance, Pacific West Bank
Scammers are becoming more sophisticated, particularly in bypassing KYC controls like liveness checks in digital onboarding—once seen as impenetrable but now easily circumvented. This shift, coupled with the volume of attacks and massive data breaches exposing confidential information, reveals that traditional defenses may no longer suffice.
As AI-driven fraud and Fincrime Grow, What is Your Biggest Concern Today?
As AI-driven fraud and financial crime continue to grow, many organizations are grappling with a range of concerns:
Lack of Resources for Generative AI in Fraud Detection
Insights from Dana Lawrence, Sr. Director of Fintech Compliance, Pacific West Bank
One of my biggest concerns about emerging generative AI is that, as I represent a smaller community bank, we may not have the resources to effectively manage the new risks it introduces. However, we still need to address these challenges. The question becomes: How do we adapt and pivot, especially given our smaller budget compared to larger regional or national banks?
Siloed Data & Poor Internal Communication
Insights from Hailey Windham, Podcast Host & CU Advocate, Bank of Fraudology
One of our main challenges is siloed data, impacting organizations of all sizes. Without better visibility into payment flows and controls, staying proactive becomes difficult. But the issue isn’t just about external data silos—it’s also driven by weak internal communication. Collaboration between payments, fraud, operations, and frontline teams is essential to tackling threats, especially as new risks continue to emerge. With transaction monitoring systems falling short, it's critical to invest in strategies that strengthen integration and infrastructure across the organization.
Tips on Fostering Awareness Around Generative AI in Fraud Detection
Insights from Hailey Windham, Podcast Host & CU Advocate, Bank of Fraudology
Financial institutions need to move beyond passive fraud prevention measures and actively build a culture of fraud awareness at every level. Here’s how organizations can strengthen their generative AI in fraud detection and prevention culture:
- Frontline Employee Training: Employees must be equipped with real-time knowledge about fraud tactics. This training should be updated regularly as AI-driven fraud tactics evolve.
- Customer Education: Customers are often the first line of defense. Financial institutions must take the lead in educating them about common fraud schemes, red flags, and steps to protect their assets.
- Two-Way Communication: Fraud prevention is a collaborative effort. Institutions should create open communication channels where customers and employees can report suspicious activity in real-time.
- Internal Incentives for Fraud Prevention: Rewarding employees who successfully identify fraud early on can be a game-changer. Fun fact: Hailey’s team used the concept of “Scooby Snacks” to incentivize fraud detection efforts—simple, effective, and highly motivating.
How to Use and Implement Generative AI in Fraud Detection
Insights from Dana Lawrence, Sr. Director of Fintech Compliance, Pacific West Bank
Start with AI Governance:
- Review the Federal Reserve’s AI Program to see an example of a regulator’s AI Governance Program.
- Having an AI governance policy is important, even if your institution isn’t currently using AI. It provides a framework to overcome indecision. Using an AI governance framework helps identify the risks and set up incremental steps for safe AI integration
- Consider the NIST AI framework to learn how to set up controls for AI.
- Refresh your knowledge on model risk management—understanding how AI outputs are calculated to ensure they are explainable.
Learn AI Use Cases in Risk and Compliance:
- Generative AI in fraud detection: Refines monitoring programs by identifying patterns in your data and suggesting new rules.
- AI agents for Level 1 reviews: AI can handle the research phase by applying set rules to data, saving time for human decision-makers to focus on final assessments.
- Anomaly detection: AI detects deviations from normal patterns using machine learning, getting more accurate as more data is processed.
- AI for research: Use AI to scrape relevant news and information, keeping your team updated.
- Auto-generating alert narratives: Tools like Unit21’s AI Agent can draft alert narratives automatically, which can then be customized and reviewed, saving time on documentation.
Prioritize Fraud Cases with AI-Driven Insights for Effective Detection
Insights from Hailey Windham, Podcast Host & CU Advocate, Bank of Fraudology
For effective fraud detection, it’s important to prioritize cases based on accurate analysis. While case scores may help, they might not capture all necessary details. An AI-powered case management system can provide a more comprehensive view.
For example, AI can flag a situation where an online account takeover is coupled with fund transfers, which requires urgent attention. This allows you to act before processing the ACH origination file, minimizing risk.
Generative AI in fraud detection also helps prioritize cases by analyzing case data and acting as a virtual assistant, triaging fraud alerts and highlighting high-risk cases that require immediate action.
Assess Workflows and Implement Risk-Managed Solutions
Insights from Matthew Hogan, Detective, Connecticut State Police
Every organization must assess its current workflow and identify where it can effectively integrate generative AI in fraud detection and prevention. AI can enhance operations in countless ways, but each institution will have unique needs and approaches. Therefore, it's crucial to adopt AI in a justified manner, implement a solid testing process, and address risk management concerns. If you're considering AI, using an established framework or borrowing one is a great way to get started.
Implement Generative AI in Your Fraud Detection with Unit21!
Generative AI in financial services is not a futuristic concept. It is shaping fraud and compliance daily, on both sides.
What has changed since this panel convened is that the question moved from whether to adopt to how much autonomy to grant. Most institutions now run AI somewhere in the workflow. The differentiator is whether they can explain what it did.
Register here to watch our webinar on “The Good, Bad, and Ugly of GenAI in Fraud & AML” and gain valuable insights and strategies in generative AI in fraud detection!
FAQs
What is generative AI in fraud detection?
Generative AI applies large language models to fraud workflows: reasoning over transaction and customer context, drafting investigation narratives, and identifying patterns that suggest new detection rules. It differs from traditional machine learning models, which produce a risk score from structured, labelled data. Most effective systems use both.
Can generative AI replace fraud analysts?
No. It removes the assembly work, gathering evidence, mapping entities, and drafting documentation, so analysts spend time on judgment instead of collection. The decision, and accountability for it, stays with a human. As Hailey Windham puts it, AI is a tool to enhance fraud fighters, not replace them.
Is using AI in fraud detection a regulatory risk?
Only if you cannot explain it. Regulators have been clear that existing law applies to AI, and the requirement is that outputs are transparent and explainable. A model that performs well but cannot be explained in terms an examiner accepts is a finding waiting to happen. Start with an AI governance policy, even before you deploy anything.
Where should an institution start with AI in fraud detection?
With governance rather than tooling. Review the Federal Reserve's AI program and the NIST AI Risk Management Framework, refresh your model risk management practices, and write the policy before you need it. Then pick one narrow use case, typically Level 1 alert research or narrative drafting, and test it properly.
How are fraudsters using generative AI?
To generate synthetic identities with convincing but fabricated data, to bypass liveness checks in digital onboarding, and to scale phishing, smishing, and pig-butchering scams far beyond what manual effort allowed. The volume and the plausibility both went up at once.
Do smaller institutions need AI in fraud detection?
They face the same fraud with fewer resources, which arguably makes the case stronger. The constraint is usually managing new risk with a smaller budget, not the technology itself. Starting with a governance framework and one well-scoped use case is more achievable than a platform-wide rollout.
Trisha Kothari is the co-founder and Chairman of Unit21, a solution that proactively mitigates risks tied to money laundering, fraud, and other illicit activities. She served as CEO from inception for over 7 years, leading the company through its early growth and evolution. Trisha is driven by a powerful mission to empower the fight against financial crimes. Her professional journey, deeply rooted in engineering and product management, includes significant roles at companies such as Google, LinkedIn, and Affirm. During her tenure as an early engineer and product manager at Affirm, Trisha gained firsthand insight into the gross inefficiency and siloed ways that AML and Fraud operated. This experience served as a catalyst for her to develop innovative AML and Fraud solutions that Unit21 now offers today.
After leaving Affirm in 2018, Trisha joined South Park Commons, a community of builders, tinkerers, and domain experts. There, she met her co-founder and began tinkering with the concept of Unit21. Follow Trisha on LinkedIn, where she usually discusses new regulatory changes to be aware of, reacts to industry trends, and shares Unit21 product enhancements.
