A recent American Banker virtual summit explored some of the hurdles to using agentic artificial intelligence. This second installment of a two-part series shares the most relevant takeaways for ACA members.
08/26/2025 12:30 P.M.
4 minute read
A panel at American Banker’s recent virtual summit, “Banking on Bots: Capturing the Transformative Effects of the Agentic AI Labor Force,” focused on ethics and the deployment of AI agents.
Panelists made it clear that while agentic AI holds great promise for banking, its adoption also presents significant challenges — from use case clarity and workflow integration to governance and compliance — that require careful oversight.
Anuj Shah, co-founder and CEO of Avaris, outlined seven key hurdles to deploying agentic AI, as well as steps organizations can take to address them.
- Use Case Clarity – Define, prioritize, and secure business sponsorship for the right use cases.
- Data Reliability – Identify golden sources of data, ensure lineage, metadata accuracy, and avoid stale or conflicting data.
- Control Boundaries – Set thresholds, role-based access, and escalation paths.
- Workflow Integration – Map how agents hand off to people, processes, or other agents, and clarify exception handling.
- Compliance & Explainability – Maintain audit trails, record keeping, and transparency to meet regulatory demands.
- Monitoring & Observability – Track performance in real time to ensure agents function as intended.
- AI Governance – Align the organization with policies, working groups, and structured review processes.
While the list may not be entirely new from an organizational strategy standpoint, it’s critical to align these considerations with existing ethics, monitoring, and best practices.
Marc Corbett, director of solutions engineering at Backbase, emphasized the importance of having a clear, strategic “North Star.”
He posed questions for financial institutions to consider: Are you optimizing customer journeys? Are you automating internal processes? What does success truly look like for your organization?
By defining the end goal, he explained, banks can then allocate resources effectively, establish the right KPIs, and work backward to build a roadmap that aligns strategy with measurable outcomes.
Governance is also important, especially knowing how, when, and why you’re deploying and what the risks are.
Ken Cluff, head of AI/ML use cases and development from Truist Financial, noted that one of their best decisions — though not fully appreciated at the time — was bringing legal and compliance partners into the process and giving them a prominent seat at the table. In fact, their AI working group now includes more privacy and legal professionals than technologists, ensuring every initiative has the right oversight from the very beginning. There must be cross-functional alignment and sign-off from key stakeholders.
Privacy and Cybersecurity
Cluff cautioned that the threat surface has expanded dramatically with the rise of agentic AI. While he stressed this isn’t cause for alarm, he emphasized the need for close integration between AI initiatives and cybersecurity teams. Agentic AI is advancing faster than almost any previous technology, and expertise in both AI and cybersecurity is difficult to come by.
“If you’re not moving forward with your cybersecurity teams, then your teams are effectively moving backward,” he said.
Corbett added that security risks extend beyond system vulnerabilities to include issues like data leakage, over-permissioning, third-party risk, and prompt injection.
He pointed to a recent example where a research paper used in model training contained hidden white text instructing AI systems to respond only with positive comments about the paper. While the immediate risk from such an instance may be low, Corbett warned that it illustrates how prompt injection could be exploited on a larger scale, with potentially serious consequences.
Building on the points about vulnerabilities and prompt injection, Brian Gannuscio, senior vice president of data and AI at Coastal Cloud, emphasized the importance of protecting sensitive data from exposure. He stressed the importance of multi-factor authentication and cautioned that sensitive information can be unintentionally exposed without proper safeguards.
Gannuscio also underscored the need for IT solutions that can mask data — even from the agents themselves — and for defining and tagging metadata tied to compliance concerns. Doing so is a strong starting point for preventing data loss, he said.
For ACA members, the lesson is clear: embracing agentic AI requires not just curiosity about its possibilities, but commitment to the governance and security practices that make those possibilities sustainable.
As the conversation around agentic AI grows, ACA will continue to share insights to help members navigate both the opportunities and the risks ahead.
Read the first installment of this series.
Remember, subscribe to ACA Daily and Member Alerts under your My ACA Assistant profile when logged in to acainternational.org.
A recent American Banker virtual summit explored some of the hurdles to using agentic artificial intelligence. This second installment of a two-part series shares the most relevant takeaways for ACA members.





