Celent estimates that AI was behind roughly 20% of the fraud perpetrated in 2024 and is poised to fuel even more fraud moving forward.
The good news is that AI models for fraud detection continue to advance, leading to higher capture rates. The next opportunity for moving the needle in fraud prevention lies in transformer technology and generative AI. Efforts by banks, card processors and their technology partners to apply generative AI to fraud operations are gaining steam. According to recent Celent research, fraud detection and prevention tops the list of functions where financial institutions are most actively exploring the use of generative AI. At the cutting edge, financial institutions and technology providers are looking to apply transformer models directly to fraud detection.
As machine learning and generative AI deliver ever more powerful solutions, the main stumbling block to better fraud prevention is not in the models themselves, but in the infrastructure they run on. Due to hardware, software, and networking constraints in large banking and payments environments, anti-fraud models typically run on only a fraction of real-time transactions, limiting their ability to detect fraud across their entire exposure.
In 2022 IBM helped solve this problem by introducing an innovative AI accelerator that ran on chips inside the mainframe environment itself. This allowed banks and processors to apply AI inferencing to all transactions in real time, greatly expanding their ability to detect and stop the fraudulent transactions running through their systems.
IBM has recently announced the second-generation of this technology. Moreover, according to IBM, their new AI accelerators can run traditional machine learning models and transformer models together, in real time, enabling financial institutions to fight fire with fire and apply the power of generative AI directly to fraud detection.
