Title: 
Using AI to fight genAI fraud at scale

Speaker:
Radu Tudoran, Microblink Ltd.

Abstract:
The rapid rise of generative artificial intelligence has brought  many opportunities, but it has also enabled fraudsters to scale their operations with unprecedented speed and sophistication. In 2024, the U.S. Federal Trade Commission received over 1.1 million reports of identity theft, and by Q3 2025 reports had already exceeded 1.15 million. American adults lost an estimated $47 billion to identity fraud and scams in 2024, an increase of $4 billion over 2023, according to Javelin Strategy & Research and AARP data.
Fraud has evolved from traditional phishing and account takeover schemes to more complex forms such as synthetic identity fraud, where criminals blend real and fabricated information (e.g., a genuine Social Security number with a fictitious name and address) to establish new, fraudulent identities. As generative AI tools become increasingly accessible, fraudsters now use them to produce convincing fake documents, simulate human-like interactions, and automate large-scale attacks. Digital document forgeries have surged, with some reports showing a 244% year-over-year increase, and deepfake-based attacks now comprise a substantial portion of biometric fraud cases, occurring as frequently as once every five minutes. These “Fraud-as-a-Service” platforms enable organized groups to rent AI tools that generate thousands of fake identities with minimal expertise.

In this talk, we will walk through how fraudsters leverage generative AI for synthetic identity creation. We will then explore defensive strategies powered by machine learning and advanced detection models designed to keep pace with increasingly adaptive adversaries. Beyond detection models, fighting AI-driven fraud at scale requires enterprise-grade software architecture. Modern fraud prevention platforms must support millions of identity verifications per day, demanding a highly scalable cloud-native infrastructure capable of rapid autoscaling, elastic compute provisioning, and seamless roll-forward and rollback deployments. These systems rely on event-driven architectures that process and monitor billions of events per month, enabling real-time insights, observability, and continuous model improvement. 

The talk will provide a practical perspective on designing such systems using cloud services, highlighting architectural trade-offs, operational challenges, and scalability considerations. By understanding both the threat landscape and the defensive toolkit, we hope this talk will serve as inspiration to join the battle to mitigate fraud in an era where AI benefits and burdens coexist.