How AI is Changing Bank Payment and Fraud Systems

Financial institutions are increasingly deploying artificial intelligence to manage complex tasks like fraud detection, credit decisions, and payment processing [1]. While these systems aim to improve efficiency and security, their growing role in automated decision-making has prompted banks to prioritize governance and human oversight to manage potential errors [1].

Key takeaways

• Nearly 90% of financial institutions are currently deploying or assessing the use of artificial intelligence [2].
• Modern AI models, such as those used by Stripe, have significantly increased fraud detection rates by analyzing billions of transactions [2].
• Agentic AI is moving beyond simple chat tools to perform complex, multi-step tasks like validating documents and recommending fraud investigation steps [1].
• Financial institutions are shifting from using fragmented, single-purpose AI models to consolidated systems trained on comprehensive transaction histories [2].
• Governance remains a primary challenge, as banks must determine which responsibilities can be delegated to software and how to maintain accountability [1].

The Shift Toward Consolidated AI Models

Banks and fintech companies are moving away from maintaining hundreds of disconnected AI systems in favor of consolidated models trained on their entire transaction history [2]. By training a single model on vast amounts of data, institutions can identify suspicious patterns more effectively, such as detecting a payment from an unfamiliar device that deviates from a customer's typical behavior [2]. For instance, the neobank Revolut developed a model called PRAGMA, which processes 40 billion transactions to handle credit decisions and fraud detection simultaneously [2]. Similarly, Stripe reported that its payments foundation model helped block nearly $112 billion in fraud last year, raising detection rates for certain fraud types on large businesses to 97% [2].

Balancing Automation and Accountability

As banks integrate "agentic" AI—software capable of executing sequences of tasks with limited human involvement—the focus has turned to the boundaries of machine authority [1]. While these agents can streamline operations by gathering information, validating documents, and summarizing findings for fraud analysts, they introduce new liabilities [1]. Executives are now grappling with how to document AI actions and ensure human intervention remains a part of the process when errors occur [1]. Because unauthorized-party fraud, such as credential theft and account takeover, now accounts for 71% of fraud incidents, institutions are under pressure to balance the speed of automated systems with the necessity of clear, measurable oversight [1].

Why it matters

The transition toward agentic AI and consolidated models marks a significant shift in how financial institutions manage risk and operational efficiency. As banks grant more authority to software, the challenge lies in maintaining governance over systems that operate with increasing autonomy. For customers, this means that while payment and fraud systems are becoming more sophisticated and faster at identifying threats, the industry is still defining the rules for how these machines are held accountable when a transaction is incorrectly flagged or blocked. The future of banking will likely depend on how well institutions can integrate these powerful tools while keeping human judgment at the center of critical financial decisions [1].