Fiserv has co-created AI agents with six banks and OpenAI

The financial industry’s race to adopt agentic AI has put pressure on the largest bank software vendors to support this activity. 

Processing Content

In response, Fiserv and FIS, providers of foundation technology underpinning most U.S. banks, have begun offering AI agents that work with their software. BMO and Amalgamated Bank are deploying a new Financial Crimes AI agent FIS recently built with Anthropic that helps investigate financial crime and money laundering. 

On Thursday, Fiserv, which has 6,000 financial institution customers (3,000 use its core systems), announced it’s launched an agentic AI operating system for AI agents that work with its core and payments software. The operating system, agentOS, was developed with OpenAI using Amazon Web Services’ Bedrock. Fiserv is also working with OpenAI on other agentic AI projects. 

Meeting bank clients’ demands

Dhivya Suryadevara, who joined Fiserv five months ago as co‑president, said she’s talked to dozens of clients about innovation and the problems they’re trying to solve.

“Everybody has the same problems,” Suryadevara told American Banker. “They have cost pressures. They want to increase their deposits. They have people retiring in their talent base. So they’re saying, with all of these pressures together, it’ll be great to get some help to deal with that.”

The bank clients also see AI as a “potential, meaningful accelerant of their business model,” Suryadevara said. But some are nervous because it’s a regulated industry, and banks need to have controls. “And on the other side of the spectrum, folks are less aware of how to deploy AI, and they would love for us to be there. We’re usually their R&D arm, and that extends to AI.”

AgentOS lets banks use Fiserv-built agents, build their own or deploy third-party agents within a controlled architecture that has “bank-grade controls.” These are controls for policy, governance, auditability and proper recordkeeping, Suryadevara said. 

“We have kill switches,” she said, for turning AI agents off when something goes wrong. “We have ‘human in the loop’ on what agents are allowed to do and where you need to involve humans. Every single thing you can think about, from a regulatory and a banking perspective, on controls, we have embedded those into the platform.”

The new AI agents

So far, Fiserv has co-developed four AI agents with six bank partners. 

One example is an AI agent for commercial loan onboarding that Fiserv co-created with First Interstate Bank in Billings, Montana, which is now piloting it. Onboarding commercial loans straddles multiple systems, “it’s super manual, and a lot of people have to spend a lot of hours on it,” Suryadevara said. 

“Delivery of an agentic operating system through our core allows us to simply integrate agentic tools into our daily workflows,” said Jim Reuter, president and CEO of First Interstate BancSystem, in a statement. The new agent has reduced manual data entry and cycle times, he said.

Boulder Dam Credit Union in Boulder City, Nevada, is piloting an AI agent for generating reports. “Today, they’re running reports which take way too long, and you have to pull data from this system, cross verify it, there’s an analyst that’s checking it, and then you generate the report, then you run the analytics,” Suryadevara said. “So there’s a lot of manual work going on.” The AI agent shortens the time needed to create reports, she said.

“As we aim to embed agentic AI into our operations, doing so within the controlled and governed architecture of Fiserv was critical to us,” said Steele Hendrix, president and CEO of Boulder Dam Credit Union, in a statement. “Our initial use of a daily operational analysis agent is helping automate manual tasks, including cutting report times down from 10 minutes to a matter of seconds.”

Fiserv is also codeveloping agents with Salem Five, City National Bank, Bank OZK and SouthState; deployments will begin this summer.

Nine software companies have signed up so far to create AI agents for agentOS: Arva, Cognext, iTuring.ai, Lumio, Osfin.ai, Sardine, Sierra, Tracfox and Trulioo.

AgentOS runs on AWS Bedrock, a service that provides a single application programming interface to access and use foundation models from AI companies like Anthropic, Meta, Mistral, Cohere and Amazon to build generative AI applications. 

“Bedrock is very flexible,” Suryadevara said. AWS has “very good enterprise grade platforms where you can access multiple LLMs. They offer the ability for us to build it the way we want to build it. We’re partnering with them to bring a secure operating system to market, and they have the building blocks to make that happen.”

Fiserv’s collaboration with OpenAI is broader. The two companies are building AI agents together. They’re working together to use AI to help with software conversions and modernization projects. Software deployments and upgrades can take thousands of hours and force banks to put other projects on hold, Suryadevara pointed out. “With AI, there’s a tremendous opportunity to collapse the effort and the timeline associated with implementations and conversions, and we’re going to work with OpenAI to say, how do we approach these in an AI-native way,” she said. 

In a third collaborative effort, the two companies will infuse AI models into existing bank products and operations. In the fourth element of the partnership, Fiserv is joining OpenAI’s Trusted Access for Cyber program. 

“That’s going to help us infuse AI into the cyber capabilities that we already offer to our banks,” she said. Suryadevara declined to say whether this is an answer to Anthropic’s Claude Mythos model, which can detect and potentially exploit software vulnerabilities faster than existing software or humans.

Fiserv may work with other AI companies in the future, she said.

“Many people are saying that AI is one of the most transformative technologies that are going to impact banking in a positive way, and I think that everybody in the ecosystem should work together to bring the right solutions in our clients’ hands,” Suryadevara said. “So we’re supportive of AI efforts in the industry across the board.” Fiserv already works with Google and Anthropic in some places, she said.

Using agentic AI with core banking software comes with a range of risks, according to Carissa Robb, managing partner at consultancy SolomonEdwards.

“The biggest risk is that agentic AI changes the scale and speed at which decisions can be made within a bank’s core system, creating new challenges around accountability, explainability and model drift,” Robb told American Banker. “A core system is the operational nerve center of a bank, so introducing AI agents that can read from and write to that environment creates a very different risk profile than traditional software. These systems can operate at a speed and scale that human oversight can’t fully monitor in real time, so the risk is often not obvious failure, but quiet failure at volume that may go unnoticed until an examiner picks up on it.”

Accountability is another issue, she said. “An AI agent may influence a decision, and a vendor may provide the model, but the bank owns the outcome from a regulatory perspective. If an institution can’t clearly explain how a decision was made or recreate how the system reached a conclusion, that quickly becomes a governance concern.”

Model drift is another risk. “AI agents adapt over time, but without ongoing monitoring and validation, outputs can gradually become inaccurate or misaligned with policy expectations,” Robb said. “And because these agents interact directly with core systems, issues in one area can create downstream impacts across deposit operations, lending, compliance, and customer-facing functions.”