The financial sector is navigating a profound shift in its relationship with artificial intelligence, moving beyond mere automation assistance to embrace agentic AI, which can independently execute complex sequences of tasks. This evolution, highlighted by recent developments from companies like Catena Labs, Primitive, and Saris, forces banks to confront a critical question: Who ultimately signs off on the machine’s decisions when the software itself is driving the process?
For years, banks have grown comfortable with automation. Software routinely reviews transactions, flags suspicious activity, routes documents, generates reports, and enhances employee efficiency. These systems primarily support human decision-making, rarely owning the decision itself. Agentic AI, however, introduces a fundamentally different proposition. Instead of merely assisting an employee, this advanced software can carry out a sequence of tasks autonomously. It can gather information from disparate systems, review documentation, complete workflow steps, escalate exceptions, and propel a process toward completion with significantly reduced human involvement.
The infrastructure for these AI agents operating within financial institutions is rapidly developing. Firms such as Catena Labs, Primitive, and Saris are focusing on embedding software deeper into processes that have historically relied on human judgment and supervision, moving beyond simple chat interfaces or productivity tools. The applications within banking are extensive and immediately apparent. Financial institutions dedicate substantial resources to managing lending documentation, conducting rigorous compliance reviews, investigating intricate fraud cases, onboarding new customers, and handling a myriad of servicing requests. Many of these activities demand employees to meticulously gather information from multiple systems, apply established rules, and coordinate work across various departments. Agentic AI promises to substantially reduce this operational burden.
While the value proposition of agentic AI is clear when viewed through the lens of productivity, the inherent governance challenge becomes equally apparent when accountability is considered. For instance, a loan officer might traditionally use software to organize a file. With agentic AI, the software could be tasked with collecting missing documents, validating information, identifying inconsistencies, requesting additional materials, and even preparing the file for a final human review. Similarly, a fraud analyst might leverage technology to identify suspicious activity. An AI agent could then assemble comprehensive account histories, cross-reference customer records, summarize findings, and recommend next steps before a human ever intervenes in the process. The delegation of such responsibilities, however, carries significant challenges and, ultimately, liabilities.
Financial institutions are grappling with these questions at a time when risk management demands are already escalating. According to PYMNTS Intelligence’s “State of Fraud and Financial Crime in the United States,” a striking 46% of financial institutions report an increasing sophistication in fraud schemes. Furthermore, nearly half of the executives surveyed by PYMNTS Intelligence and Block cite regulatory pressures as a major challenge, while 41% point to pressures associated with faster and more diverse payment systems. In response to these growing threats, 68% of institutions have increased spending on fraud detection capabilities. These figures underscore why discussions surrounding agentic AI frequently pivot to governance rather than solely focusing on its technical capabilities.
The emphasis on control is paramount. As noted in announcements from AI firms, Primitive, for example, has specifically highlighted “controls, measurement and oversight” as central components for deploying AI agents within regulated environments. This reflects a fundamental reality for every bank executive: operational authority cannot simply be transferred to software without establishing clear boundaries around what the software is permitted to do and how its actions are rigorously monitored. The complexity is further compounded by the persistent threat of fraud. PYMNTS Intelligence data reveals that unauthorized-party fraud now accounts for a substantial 71% of fraud incidents and losses, largely driven by credential theft and account takeover activity. Such events not only result in direct financial losses but also inflict damage to customer loyalty, reputational harm, and lost business opportunities.
The embrace of agentic AI shines a direct spotlight on the question of authority. Banks must meticulously define which actions an AI agent can initiate autonomously and which decisions necessitate explicit human approval. Critical considerations include how exceptions are handled, how all agent actions are comprehensively documented, and, most importantly, who bears accountability when an AI agent makes an error that impacts a customer, a transaction, or a regulatory obligation. These intricate questions reside squarely at the intersection of advanced technology and robust governance frameworks. For years, financial institutions have focused on determining the capabilities of machines. The discussion has now decisively shifted to the extent of authority they are prepared to grant these increasingly capable systems.
As agentic AI moves from concept to operational reality within the banking sector, the industry’s ability to balance innovation with stringent control will define its success. The ongoing dialogue will not merely be about technological advancement, but fundamentally about redefining human-machine collaboration and the ultimate locus of responsibility in an increasingly automated financial world.
