Griffin, a UK-based banking-as-a-service (BaaS) fintech, recently announced plans to integrate autonomous AI agents into its core banking infrastructure. The prospect of machines running your bank may at first sound like a nightmare, but these aren’t the worthless, customer service chatbots we now encounter every place we might even consider transacting online. True agentic AI offers a bounty of never-before-possible opportunities and risks.

What is agentic AI?

Agentic AI refers to artificial intelligence that operates autonomously, enabling it to make decisions and perform tasks without human involvement. In the banking sector, this means AI agents could feasibly handle complex financial crime tasks, detect and act on fraud risks, manage credit risks, and resolve customer service inquiries. 

Griffin's announcement that it intends to offer a Model Context Protocol (MCP) Server is significant because it will facilitate interactions between its core banking systems and customers’ AI models.

What is a Model Context Protocol (MCP) Server?

Put simply, an MCP Server provides AI agents with relevant data, rules, and business context.
It enables such agents to act consistently and intelligently over time. Within a fraud context, it also supports the agent in ensuring that past outcomes influence the decisions the agent makes in the present. The MCP Server also contains AI agents, ensuring they operate within defined limits.

An MCP server in an agentic bank would coordinate multi-step actions such as opening an account, performing know-your-customer (KYC) checks, and making decisions based on responses from multiple systems.

The MCP server is the infrastructure that enables reliable and explainable decision-making by AI agents. It ensures they can understand nuanced contexts (e.g. transactional patterns), remain within policy, and, most importantly, act transparently.

What's in it for the bank and its customers?

Agentic AI has the potential to power hyper-personalisation, anticipating customer needs by suggesting financial services products that fit with the customer's circumstances.

Are you sick of paying rent? No problem! Let me open an individual savings account for you and provide you with a mortgage offer once you can meet the deposit requirements.

The AI agent might also remind you to curtail your spending so you can meet that objective, assuming, of course, it agrees that skipping brunch is the path to homeownership. Unlike the dumb chatbots we've all endured, AI agents will also be able to hold multi-step, context-aware conversations across multiple channels (chat, voice, app., etc.).

From the bank's perspective, agentic AI offers the opportunity to achieve significant efficiencies by automating ID checks, KYC checks, and risk management. AI agents should also excel at tasks that often involve humans papering over the cracks in fragmented systems, seamlessly addressing tasks such as reconciling fraud alerts in a manner that is fully compliant with the bank’s policy framework.

Regarding fraud specifically, agentic AI provides the capability to detect subtle, evolving fraud patterns using both historical and real-time context. Importantly, it should also be able to do so in a manner that is explainable and defensible from a regulatory perspective.

What risks does agentic AI pose?

Much like humans, AI agents will not be infallible. When misconfigured or poorly monitored, fraudsters (or their own AI agents) could trick these digital bank employees into approving fraudulent transactions, unwittingly exposing sensitive data, or exposing vulnerabilities to hijacking through adversarial inputs.

Pioneers such as Griffin are right to highlight the opportunities offered by an agentic bank. While the benefits are substantial, benefiting banks and their customers, we can't take our eye off the ball. The legal, regulatory, and financial crime risks will require new tech-centric approaches to governance and compliance.

We should also never forget that while the sector will seek to deploy the technology responsibly, it is inevitable that an agentic bank will find itself going toe-to-toe against an agentic fraudster, money launderer, terrorist, or human trafficker. These criminal agents will be unconstrained, empowered, and ruthless. While these issues are not insurmountable, striking a balance between risk and opportunity will be essential.

One thing is certain: The pace of innovation will quicken, and only those who stay ahead of the risk will prosper.