GS interviews Treasury Prime's Chris Dean

As artificial intelligence moves deeper into financial services, banks and fintechs are grappling with how to harness its potential without compromising governance, transparency and regulatory compliance. In this Q&A, Chris Dean, co-founder and CEO of Treasury Prime, discusses where AI is delivering real value today, why legacy infrastructure poses significant challenges, and what institutions must do to build systems capable of supporting autonomous decisioning while maintaining control in a highly regulated environment.

Green Sheet: Many predictions position banking as a major frontier for AI-powered services. What capabilities are driving that expectation right now?

Chris Dean: That potential to make core banking operations faster and cheaper is what is driving the predictions. Banks that can securely process transactions kicked off by AI agents or merchant-side systems will win the volume as demand grows. Inside the bank, AI tools can absorb the routine stuff: coordinating onboarding workflows, assembling regulatory reporting materials, and reconciling data across systems.

That frees up teams to focus on work that actually requires judgment. But banking is banking. Strict regulatory requirements mean automation has to be rolled out carefully, with real governance and oversight baked in from the start.

GS: You've raised concerns about whether today's banking infrastructure is ready for autonomous AI decision-making. Where do you see the biggest gaps?

CD: There is no room for hallucinations here, and most banking systems were never built to catch them. They were designed for human operators, not AI agents, so the infrastructure for monitoring, governance, and traceability around AI actions simply does not exist yet.

AI is already pulling its weight on the repetitive stuff, automating manual tasks and helping teams sort through mountains of work. But operating independently is a much higher bar. You need infrastructure that shows what happened, what data informed the decision, and how a human can review or intervene.

Without that, you are trusting a system that sometimes makes things up to operate inside one of the most tightly regulated industries on the planet.

GS: When financial institutions attempt to layer AI on top of legacy systems, what kinds of risks can emerge?

CD: The biggest risk is that AI ends up operating in an environment it wasn't designed for. Legacy systems often have fragmented data, inconsistent workflows and limited visibility across processes. If AI is acting on incomplete or poorly governed data, mistakes can happen quickly, and it becomes harder for institutions to detect or explain them.

Another risk is that AI can scale problems just as easily as it scales efficiencies. If the underlying data in legacy systems is inconsistent or poorly structured, AI can amplify those issues across decisions, transactions or customer interactions before anyone realizes something is wrong.

There's also the challenge of auditability; banks need to be able to clearly explain how decisions are made, and that becomes difficult if AI systems are layered onto infrastructure that wasn't built with transparency or real-time oversight in mind.

Finally, integrating AI into older environments can create new security and data-governance risks if sensitive financial data isn't properly isolated and controlled.

GS: You've said AI-ready infrastructure must be "native rather than bolted on." What does that look like in practical terms?

CD: In practice, it means AI activity is built directly into the system's controls and oversight. The infrastructure can track when AI is making a recommendation versus taking an action, log those decisions, and allow institutions to review or intervene. When AI is simply layered on top, that visibility and control often get lost.

It also means the infrastructure is designed to treat AI like any other critical system component, with clear permissions and guardrails around what it's allowed to do. When AI is native to the platform, banks can control which systems it can access, what data it can use and when human approval is required.

That kind of built-in governance is much harder to achieve when AI is bolted onto legacy infrastructure that was never designed to support it.

GS: How could getting the underlying infrastructure wrong create compliance or regulatory challenges for institutions deploying AI-driven services?

CD: Regulators expect financial institutions to understand and supervise how decisions are made. If AI is initiating transactions or influencing approvals, banks need clear records of what the system did and why.

Financial institutions already operate within well-established risk and governance structures, so the challenge is integrating AI into those existing frameworks rather than creating entirely new ones. With strong oversight built in from the start, banks and fintechs can continue to move quickly while maintaining control.

Without infrastructure that captures and governs those actions, institutions may struggle to demonstrate proper oversight or accountability.

GS: As banks move toward more autonomous transactions and decisioning, what steps should they take now to build a foundation that can support long-term innovation?

CD: Banks need to start by focusing on governance and visibility, not just the AI models themselves. Assuming these systems can simply run on their own once deployed is unrealistic and could put sensitive financial data or critical decisions at risk.

In practice, AI still depends on strong data, clear guardrails and ongoing human review. Over the next few years, AI will help banks move faster and handle operational work more efficiently, but accountability and regulatory responsibility will still sit with humans. 

Notice to readers: These are archived articles. Contact information, links and other details may be out of date. We regret any inconvenience.