Artificial intelligence

OpenAI CEO and co-founder Sam Altman was taken aback by the financial sector’s early interest in ChatGPT and his firm’s other generative artificial intelligence products.

“It's gotten much better, but when we first launched, the thing that the average person thought about AI is that it hallucinated a lot,” he said during an on-stage chat in July with Federal Reserve Vice Chair for Supervision Michelle Bowman at a capital framework conference hosted by the central bank.

Although Altman pegged ChatGPT’s hallucinations at 0.1% or less, “that branding kind of stuck with us for a while, so we assumed that financial services … were not gonna be our early adopters,” he said. But some of OpenAI’s biggest early enterprise partners were financial institutions.

“Morgan Stanley, Bank of New York, these are major partners of ours that we're doing fantastic work with. And we were kind of like, ‘Are y’all sure?’ and they were like, ‘Yeah, we really want to do this,’” he recalled.

These institutions and others figured out how to use the technology, and “how to structure it enough that they can rely on it” for critical processes, he said.

“The thing that someone said that stuck with me is, ‘Hey, we realize this is a new technology and we've got to put some new controls on it. But the risk of us not doing this … is that we don't continue to exist as a business,’” he said.

Since Altman and his OpenAI co-founders launched ChatGPT in November 2022, banks have harnessed generative AI to increase productivity and efficiency and to manage vast quantities of customer data.

Generative AI should be able to handle up to 40% of daily tasks by year’s end, according to bank executives surveyed by KPMG in April, and 57% said generative AI is an integral part of their long-term vision for innovation and remaining relevant.

Due to AI, the cost of each unit of intelligence has been driven down “by more than a factor of 10 each year for the last five years,” and Altman expects that to continue for at least the next five years.

He pointed to the saying "Electricity too cheap to meter," adding, “We didn't quite deliver on that as society [but it does] look like we're about to deliver on ‘intelligence too cheap to meter.”

The AI boom has been compared to the industrial revolution. JPMorgan Chase CEO Jamie Dimon said it could be as transformative as electricity or the internet. But Altman, in at least one way, implied it’s bigger.

“The internet did change a lot of things, but I don't ever remember something that would have cost $10,000 of knowledge work a year ago, costing $1 or 10 cents, or whatever it would cost now. This is just unprecedented,” he said.

One thing he did compare AI to, though, was the transistor – a 1947 technological innovation that is now a fundamental part of nearly all modern electronic devices.

“We don't think about this as a transistor device or that as a transistor device – we think of this as like a microphone and a computer and a screen and a camera and whatever else – but this technology was an incredible discovery,” Altman said. “It changed what we were able to build. It became part of kind of everything pretty quickly.

“You don't talk about companies much as transistor companies … in the same way, I don't think you'll be talking about AI companies for very long,” he said. “You will just expect products and services to use this technology. You won't think about it that much. You will expect products and services you use to be smarter than you, and that will be a normal part of the world, like the transistor.”

But with innovation comes risk. Altman wasn’t shy about addressing how AI has created opportunities for fraud. He’s “terrified” that there are some banks that use voice recognition technology to verify identity, he said, because “that is a crazy thing to still be doing; like AI has fully defeated that.”

He was referring to deepfake attacks, where cybercriminals use generative AI to copy a person’s voice or image and then use that voice or image to commit fraud, which a 2024 business.com survey found has affected one in 10 companies.

In the face of deepfake attacks, banks must fight fire with fire and invest more in artificial intelligence themselves, Fed Gov. Michael Barr said in April.

The ways that bad actors use deepfakes “is going to get so compelling,” Altman warned, that “people are gonna have to change the way they interact.”

“They're gonna have to change the way they verify,” he said. “Like this person calling me right now. It's a voice call. Soon, it's gonna be a video FaceTime that’s indistinguishable from reality. Teaching people how to authenticate in a world like that, how to think about the fraud impact – this is a huge deal.”

Many banks may be eager to weave artificial intelligence into their operations and quickly uncover efficiencies, but TD’s Ted Paris is prioritizing a slower approach that hews closely to what customers want.

Paris, TD Bank’s head of analytics, intelligence and artificial intelligence, said the lender is employing AI in a number of areas, including in fraud defense, conversation summaries and a newly added virtual assistant for TD Securities employees. The Toronto-based bank, which has U.S. headquarters in Cherry Hill, New Jersey, is also exploring uses related to content creation and enabling customer self-service.

In the U.S., TD is “heavily oriented around inwardly-, colleague-facing efficiency and productivity plays from the genAI space,” he said in a recent interview, “that have human-in-the-loop-type of methodologies or approaches.” 

That allows TD to ensure such activities, particularly in the genAI space, fit within its risk framework, Paris said. 

“It has plenty of space to make lots of improvements and a big impact, but also ensures that we're not disconnected from the human connection,” he said. “We’ve been on the more cautious end. Trust is going to be important.” 

Editor’s note: This interview has been edited for clarity and brevity. 

BANKING DIVE: What are the biggest challenges with scaling AI adoption across the bank?

TED PARIS: I think about it from two perspectives. How do we think about challenges from the clients’ receptivity? And how do we think about it internally, in terms of what we're rolling out, and what spaces we want to consider? And then I’d add a third one – the talent space as well. 

With the client piece: it’s something people are beginning to feel confident about, they’re curious, they’re beginning to feel comfortable in many regards. However, they also have some concerns with respect to, is this going to disintermediate me from the interaction I want to have with a person? Especially around things like retirement planning, investments, they want somebody to confirm for them, “Yeah, that's the way they interpret it. That's what's going on.” We’ve got to be able to meet the client where they are in the journey. 

On the talent side, there’s a lot more competition for the talent in the space, and you’re going to need the right talent to be able to deploy these capabilities effectively. Just as much as there's competition for the data scientists, there's also competition for and a need for the rest of the professional ecosystem, the specialists that go around it. So people who help with model risk management, change management, legal and compliance experts, the non-operational risk people. The opportunity this presents can transform an organization's operations and servicing environment in the way in which it takes a lot of change management to make sure you're going to be successful, a lot of understanding of the associated risks.

Is it a struggle to find talent in this realm? 

No, I wouldn’t say we’re struggling, but it's clearly a different dimension. I've been in sessions with my peers, and we all have said: You know what? This is something you can't think about as we may have in the past, where those in charge of the data and analytics space were focused on the data scientists, data engineers. Now, there’s bigger impacts. More broadly, it’s important to say we’ve got to think about the entire ecosystem to be effective.

Do you think that will cancel out the workforce reductions some banks are touting as they implement technology to a greater extent across their organizations? 

It’s a different conversation as to what impact it has on the workplace, in terms of job replacement. Independent of that, you still have the need to have a better understanding of the nature of the risk that this capability creates. You still have the need to be able to deliver what are inevitably going to be larger-scale programs that will have change management implications, that are larger scale than what traditional analytics and data projects might have otherwise had. That does create different types of opportunities, whether it’s for purposes of deploying or implementing projects, or the outcome of what those projects then look for. 

What’s a notable finding from TD’s consumer survey in June on AI comfort and trust?  

There's lots of confidence in what these tools can provide. But when it comes to interacting with it, there are instances in which customers have said, you know what, I'd rather have this conversation with the chatbot, because it's something that I'm just a little bit embarrassed about. So there are certain spaces in which they’d rather have the conversation without a human in the loop, when they want to be able to ask and get feedback in a more anonymous manner. So there are two sides to that coin. 

What’s a change on the horizon?

Our virtual assistant is currently heavily rules-based. Over time, in the industry, you’ll see them evolve to be more AI-driven. 

Many out there today are rules-based, with a pretty established set of questions and answers that we'd expect to come up. And you can write something in, and it'll interpret the question. It'll have a very structured, pre-designed answer. Where the space is looking to go is in one that is more AI-driven, and that potentially those answers won’t be limited to some pre-designed form of responses. When you do that, just like ChatGPT, you'll have a broad array of responses that could come out. And that’s where you have to make sure you're managing the associated risks.

Sponsored content

By Workday

AI is fueling significant change in the financial services sector, with 81% of companies having invested in it.

But organizations are struggling to keep up with some of the regulation and legislation that comes with it. Recent research shows that only 9% of UK leaders say their companies are prepared for incoming regulation, and 14% do not have an AI regulatory risk framework in place.

“The industry is spending a large amount of money exploring AI, but a lot of businesses are struggling with how to deploy it,” says John Davison, CIO at insurance and technology firm First Central Group. “Fundamentally, it comes down to risk.”

Workday recently organized a roundtable discussion to discover which strategies industry leaders are putting in place to counteract these fears and drive productivity.

AI is a vital timesaver.

Financial services firms are already saving significant time with AI. Dan Godfrey, group people transformation and talent director at Aviva, explains how AI is helping the insurer process claims more quickly: “We can summarize information a lot more easily and then focus on the customer conversation rather than having to look through multiple databases.”

Another example of where AI has significantly sped up processes is in confirming proof of address. This is saving hours of employee time, as well as improving fraud detection. “AI is great at pulling text from a document, comparing that text to the information we already have on file, and spotting whether or not it’s been tampered with— something that can be really hard to do by hand,” Davison explains.

These changes are also improving the employee experience. “The agent experience is enriched because they’re no longer having to do those really mundane, repetitive tasks, looking at document after document,” adds Davison. “That’s a really good example of the win-win that you get from AI.”

In addition to saving time on repetitive tasks, AI is helping employees plan new career paths and upskill for the future. “We can ask them to pick the role that they’re aiming for, then AI carries out an analysis of the gap between their skills today and the skills they’ll need to attain and fulfill that role, and provides bespoke learning pathways.”

Balancing opportunity with risk.

“The financial services industry is creating efficient operations and reducing costs with AI,” confirms Clare Hickie, CTO, EMEA at Workday. Despite these productivity gains, however, financial services leaders know they must proceed with caution.

“The balancing act is that AI is moving really fast, and there is an opportunity cost to any kind of delay,” says Davison. “So, you’ve got to balance your risk and reward.”

Businesses need to walk this narrow bridge to successful AI deployment with care. AI can impress customers, but it also carries the potent risk of potential misinformation. Back-office applications can represent lower-risk opportunities to build knowledge and confidence.

Godfrey says building a strong governance strategy is essential for financial services organizations using AI. “The big considerations really sit around data privacy, governance, and ethics. We’ve got clear governance in place, and we also ensure that there is a human in the loop.”

AI agents bring new possibilities in autonomous technology.

Organizations are beginning to invest in AI agents— systems trained to act on behalf of an enterprise or individual, performing tasks and interacting with data or other systems autonomously.

Agentic AI has enormous potential to accelerate data entry and make processes such as customer onboarding and dispute resolution more efficient. First Central Group is excited about its customer service potential. “Imagine you want to make a claim and you go to a chatbot,” explains Davison. “A traditional AI model would provide pathways for a customer based on a predetermined set of conditions. But what if your contact center is busy, or you can’t answer the phone at that moment?”

AI is already proving invaluable in financial services, but if organizations want to be successful in the next exciting stage of AI implementation, they must strike the perfect trade-off between ambition and vigilance.

“Organizations that are too cautious are going to get left behind, and organizations that are too open are going to trip themselves up,” says Davison. “The ones that are going to succeed are the ones that are able to balance risk and reward.”

Find out more about how to deliver value with Workday Illuminate™.

---

This content was paid for by Workday and produced in partnership with the Financial Times Commercial department.

First published at https://www.ft.com/partnercontent/workday/ balancing-risk-with-reward-the-financial-services-ai-success- story.html

Reprinted with permission. FT and ‘Financial Times’ are trademarks of The Financial Times Ltd.

Since the July rollout of Ally’s proprietary artificial intelligence platform, the breadth of use is what’s surprised Sathish Muthukrishnan, the bank’s chief information, data and digital officer.

“We have people in the sales force that are using it, people in the operations side, customer care associates using it; obviously, folks in the technology side; marketing; our risk control partners, risk compliance; audit, privacy – they're all big users of it,” said Muthukrishnan, who’s been in his role at the digital bank since 2019.

The Detroit-based lender gave its 10,000 employees access to Ally.ai in July, after testing it with a smaller group for more than a year. About 400,000 prompts have been submitted to the platform, and adoption is at about 50%. 

The bank wants employees to use the platform, which was built in-house, to handle tasks such as drafting emails and proofreading copy, to free up their time for other projects. 

When asked how AI might affect the company’s headcount, Muthukrishnan said it’s set to “have a meaningful impact on the business outcomes.”

Ally has “ambitious” growth plans, so for the company to generate more revenue while maintaining current spending levels, “technology and AI become critical,” Muthukrishnan said in a recent interview with Banking Dive. “That's both driving efficiency and effectiveness. It's not just efficiency of cost; it's efficiency of speed.” 

Editor’s note: This interview has been edited for clarity and brevity.

BANKING DIVE: Where does Ally go from here with AI?

SATHISH MUTHUKRISHNAN: Since the launch, there is tremendous demand and a lot of use cases coming our way. Now, let's turn the tables and see how we can identify use cases that are harder to solve on the business side, and how do we bring that to the forefront? 

With the pace at which technology is evolving, something that seems impossible, something that seems super hard to solve right now, we will be able to solve in a few months. So we want to tackle those hard problems now, and we want to do it collectively across the organization. 

Our CEO has asked me to come and educate the entire executive committee on how we are advancing in AI, and we're going to call it an executive committee AI day, and it's just purely to set aside dedicated time, bring us all together, fully focused on AI. These are all busy people running big organizations, so there's a little bit of pressure on making sure that I use their time efficiently. But we're going to talk about what are the things that we can collectively solve for the company. We have thoughtfully rolled out AI, and there is interest across the company, but we need to bring the company along.

How has Ally’s AI governance approach evolved since implementation?

It might sound like a cliche, but we focus on doing simple things savagely well. Things that are simple – having risk controls, having data protection, having access controls – can be cast aside because you see the shinier object. 

For us, to have an AI working group, then having an AI governance steering council, then having an enterprise-level committee, then the board – having this many levels of governance to ensure that AI is scaled safely and responsibly is super critical. We did the hard work ahead of time, we have exercised this governance muscle extremely well, and people have gotten used to it.

How do you see the role of AI agents evolving at Ally in the coming years?

Agentic AI allows you to look at the complicated paths, complicated processes, and allows you to digitize that. It’s still in an experimental stage for us. 

For example, all applications in our tech ecosystem have observability. If there is an issue, we want to be the first to find out, before the customer finds out, or our business partner finds out. So a ton of alerts come our way. If I have to process those alerts, but not increase my headcount as I'm increasing the number of customers, I'm looking at agentic AI to do that. The usage of digital has doubled in the last four years by our customers, but the cost of serving them has gone down. That's because of the introduction of new technology. 

If you want somebody to reset your password, that could be agentic AI that does that internally. Those are some of the experiments that we are doing; nothing that is in production or at scale yet.

What are your biggest concerns, as the bank’s tech chief?

No. 1 is cyber. It’s top of mind. Not just because of what is happening today, but how easy it is for cyber criminals to get access to technology and create more sophisticated cyber attacks. But as a regulated institution, you have to watch how technology is being adopted.

No. 2 is the pace at which the technology is changing. It weighs on me, because it's a constant learning process for the organization, on top of delivering at pace. So your learning agility and your execution agility always has to be, like, in the fifth gear, and that's hard. And these cyber events are like sharp turns or speed bumps on the road while you're running a marathon at sprint speed.

And No. 3 is, how do you educate and bring the organization along? Technology or AI is not here to replace anybody; it is to make sure that you take away the mundane part of the job and make them even more effective, so they can serve our customers better. So how do you do that, how do you bring everybody along? And that is a patience game. But at the same time, the speed at which the technology is changing, you can’t be patient. So it's a partnership with them, to bring them along.

How do you think about the future of your role?

Today, everybody can be a CIO because of ChatGPT. You would walk into a room and you felt like you had knowledge that nobody else did. But as you're talking, people can be on their phone verifying your answer or getting a better answer, because ChatGPT is the better CIO, whether we like it or not. 

On the other hand, thinking about the art of the possible, using technology as a tool to re-underwrite, re-imagine your business and asking the difficult questions of, are you serving your purpose? Are you serving your customers right? Do you have the right talent and capabilities to serve the customers? Can you do it differently? Can you do it faster? Can you do it with better technology? Can you enhance the user experience?

It’s up to the technology leader to figure out, where do you want to be? Do you want to be in a defensive mode, or do you want to be in an offensive mode, partnering with business leaders and figuring out, what is the next version of the product going to be?

As large companies strive to employ artificial intelligence to bolster efficiency and growth, Citi’s Tim Ryan believes getting employees to buy into an AI strategy is even more important than the right technology choices.

“There's no way we can sit in New York and come up with the strategy for a global company and implement it if we're not bringing our people on the strategy,” Ryan, head of technology and business enablement at the bank, said at the Money20/20 conference in Las Vegas in October.

AI is “a standing topic” with Citi’s board, noted Ryan, who joined the bank in 2024. He had been a senior partner at PricewaterhouseCoopers when he was tapped to lead Citi’s efforts to transform its operations and adopt emerging technologies with efficiency in mind. 

About 180,000 of the New York City-based bank’s 230,000 employees are using its AI tools, Ryan said. Staffers wrote 7 million prompts in the first eight months of this year, and have done about 1 million automated code reviews that have freed up 100,000 hours each week, he said.

“Companies that are our scale, they will win because of how they bring their people on this journey,” he said. “That's what will determine the winners and the losers. It won't be somebody standing on this stage saying, ‘I should have chosen this tech stack versus that tech stack.’ It won’t happen. What’ll happen is, it'll be, ‘I could have done a better job getting our people to buy into the strategy, equipping them to implement the strategy we're designing.’” 

That approach creates what Ryan called “an innovation dividend,” where employees then spot new ways to serve customers, drive revenue growth, manage risk and operate more efficiently.

“It's the single biggest reason we have momentum,” Ryan said. 

In his experience working with CEOs while at PwC, Ryan said change management can be underestimated in optimizing a certain strategy.

“Don't just sell tech,” Ryan told conference attendees. “You have amazing technology. When you're talking to people like me or my peers, think about, how does it get implemented at the scale of the organization? 

“Sell tech and people, and that will win,” Ryan said.

AI ‘accelerators’

Citi’s 4,000 “AI accelerators” rank as a key element of the bank’s change management approach, Ryan said. Those are staffers across all five of the bank’s businesses and various functions who’ve “leaned in” to the bank’s strategy around AI, he said.  

“These are the people that we meet with on a regular basis, to understand what's working in the tools, what's not working in the tools, what systems do we need to integrate with, what the next release of our tools needs to bring,” Ryan said. 

In conjunction with that, it’s crucial for Ryan to “have a leadership team around me that really embraces the concept of listening,” he said, “which is easy to say and hard to do.”

Ryan also noted the importance of the bank’s cloud-first infrastructure in enabling innovation, especially when it comes to agentic AI. 

The bank employs a hybrid cloud strategy, investing both internally and working with partners such as Google, Anthropic and Cognition Labs. 

“Our cloud strategy allows us to bring in the right parties and cohesively get that scale,” Ryan said.

One of the biggest challenges is to speed it up safely, he said. 

“If it takes 12 days to get something tested, I think it’s too late,” Ryan said. 

The lender has made adjustments over the past two years to bump up its testing time frame, from days to hours, “and that agility is incredibly important.” The bank is also able to deploy changes faster, he said.

In risk, the bank is using AI agents and working to scale that effort, “but not taking that human out of the loop,” he said. That offers the opportunity “to give our stakeholders more confidence around risk and controls.” 

Citi is also eyeing opportunities to use AI agents in its wealth and retail banking businesses, he said.

The bank recently rolled out agentic AI capabilities to 5,000 employees using its proprietary AI platform. Workers issue a prompt, and an agent can take multistep directions and complete tasks quickly. “We will scale that going into next year,” Ryan said.

There was no shortage of AI-related discussion at Money20/20, including its effect on jobs in financial services. 

E-commerce giant Amazon in October said it was cutting about 14,000 workers as it invests further in AI. Top banks such as JPMorgan Chase, Bank of America and Goldman Sachs have been eager to highlight how they’re employing AI; at the same time, the companies are cutting workers or slowing hiring. Citi, for its part, has trimmed its headcount as part of a broader effort announced in 2024 to cut 20,000 employees by 2026.

Ryan, however, emphasized the importance of employees’ role in carrying out the bank’s AI strategy. 

“Our job as leaders, regardless of whether you're in a fintech or a large financial organization like Citi, our job as leaders is to bring our people along, not to scare them,” Ryan said.

With artificial intelligence doubling its computing power every 100 days, according to the World Economic Forum, the biggest hindrance to AI adoption isn’t limits to its capabilities, BNY CEO Robin Vince said. 

“It’s actually going to be human adoption and culture and potentially organizational inertia that will get in the way,” Vince said in October during a fireside chat at the Money20/20 conference in Las Vegas.

“My mantra on this is, AI is for everyone, everywhere, everything at BNY,” he said.

The bank has been big on AI adoption, launching its proprietary AI platform Eliza in 2024 to improve employee workflows and client service, and allowing employees to create AI agents for specific tasks. At present, around 15,000 employees have built their own agents, and the bank is in the process of building some 150 AI-powered solutions to solve problems throughout the organization.

The bank recently began offering AI training to community bankers, too.

BNY has a history of embracing innovation, Vince said: The firm got its first computer 70 years ago and started doing business primarily via computer 50 years ago, he said. Thirty years ago, BNY began using the internet to process transactions.

“If you stick your head in the sand and ignore what’s going on around you with the technology of the day, then you’re going to become obsolete very quickly, and history is littered with firms of all industries that are living examples of that,” he said.

Around 1,000 BNY employees have received more than 40 hours of training in AI. As employees use it more, they “stretch the boundaries” of what they thought was possible, Vince noted.

“On day one, we do a few simple ChatGPT queries, but then before you know it, as we were talking about, you can actually be pretty thorough in the way that you prompt and interact and actually accomplish a mission with AI in your own life,” he said. “Then they actually can be creative with it. How can we solve a customer problem? How can we deploy AI in a new way? How can we hire a new digital employee to perform some new set of tasks? And that gets very exciting for people.”

“They can create, in our opinion, a flywheel of excitement associated with their AI adoption at the company, and then feed it with training and tools. It's a very powerful thing to be able to do,” Vince said.

So, too, has BNY embraced digital assets. It became the first U.S.-based global systemically important bank to custody crypto in 2022, and now custodies three coins (Circle’s USDC, SocGen’s CoinVertible and Ripple’s RLUSD).

Vince said he has a “very rudimentary perspective on that, which is, if there's going to be more stuff in the world, we want to look after it,” he said. “We are the world's largest looker-after of stuff.”

BNY looks at blockchain and distributed ledger technology as “an opportunity to be able to reinvent … the processes that underpin the financial system,” Vince said.

When there are developments that allow money to move faster and more efficiently, liquidity is created, and higher capacity attracts new participants in the market, he said.

Artificial intelligence is expected to drive up to 20% in net cost reductions for banks as the technology is implemented across the industry, according to McKinsey & Co.’s Global Banking Annual Review 2025 published in October. But AI-driven cost savings aren’t expected to last, as competition will erode gains for banks and savings will pass to customers rather than the industry. 

The banking industry’s net income reached $1.2 trillion in 2024, the highest total among every other industry, the report found. However, valuation of the banking sector trails other industries by roughly 70% as favorable conditions are expected to change for the industry, the report found. 

Despite potential productivity gains from technology adoption, the sector will grapple with “disruption stemming from advancements in AI, intensifying competition from nonbank providers such as fintechs, and evolving customer expectations,” the report said. 

AI promises to be a powerful force for banks’ productivity gains, but to capture value, banks will have to transform the way they operate and prepare for technology-driven change.  

The extent of AI’s impact on the banking industry revolves around banks’ ability to become fully agentic and lower operating costs, as well as customers’ adoption of AI to manage finances, according to the McKinsey report. 

Agentic AI is poised to have the most significant impact on banks’ operations, the report said. Leading financial firms are investing heavily in AI agents, with banks such as BNY, Capital One and JPMorgan Chase building architecture for agentic AI workflows. 

The banking industry likely will see the emergence of a collaborative model in the near future, with one human employee supervising 20 to 30 AI agents autonomously managing complex end-to-end workflows, the report found. BNY deploys 117 agentic AI tools for managing different aspects of the banks’ operations, Leigh-Ann Russell, the bank’s CIO and global head of engineering, shared at the Gartner IT Symposium/Xpo in Orlando, Florida in October.

Despite its promises, AI will prove to be a double-edged sword for the banking industry, according to the report. Staying anchored in tangible outcomes will be a significant focus for players like JPMorgan.

While there are “productivity tailwinds” from AI, the company plans to balance its approach by not rushing into a hiring spree for building out AI, CFO Jeremy Barnum said during a quarterly earnings call in October.

Additionally, requiring proof of AI cost savings is a “very hard thing to do,” he said.

“Hard to prove and might, at the margin, result in people scrambling around to use AI in ways that are actually not efficient and that distract you from doing underlying process reengineering that you need to do,” he said.

Big banks are reaping the rewards of growing investment in artificial intelligence research, according to an Evident Insights study published in September. The research firm tracks AI adoption across 50 of the largest global financial institutions.

The volume of research has surged as the scope has narrowed to applications, the study found. The number of published papers grew sevenfold from 2019 to 2024 among the banks analyzed, surpassing 750 in 2024, while the focus shifted to agentic AI and away from computer vision, scientific discovery and biomedicine.

“This isn’t research for research’s sake: it’s laying the foundation for faster deployments, smarter trading agents, and the next frontier of AI-driven financial services,” Alexandra Mousavizadeh, co-founder and co-CEO of Evident, said in a release accompanying the report. “The leading banks are pushing the frontier on emerging technologies like agentic AI – building the architectures and workflows that will soon underpin real-world applications.”

Financial firms invested heavily in tech talent and information technology modernization to navigate generative AI adoption roadblocks in the heavily regulated sector, cutting through a haze of uncertainty around model accuracy concerns that clouded early efforts.

JPMorgan Chase, Wells Fargo and Citi led the talent recruitment charge earlier in 2025, as banks sought out AI model development, platform engineering and project management technicians to activate use cases. Concurrently, firms increased spending on cloud, cybersecurity and data technologies to support AI adoption, according to a 2025 Broadridge Financial Solutions survey of more than 500 banking technology and operations leaders.

Pure and applied research proved to be a crucial adoption lever among leading banks, too, driving deployments of time-saving coding assistants, knowledge management tools, and customer service chatbots. 

Evident’s analysis spotlighted five banks that were responsible for two-thirds of published AI research in the sector last year. JPMorgan Chase led the pack with 37% of publications, followed by Capital One at 14%. Wells Fargo and Royal Bank of Canda each held a 5% share of publications, and TD was fifth with 4%.

As research papers proliferated, the number of use cases introduced by banks tracked by Evident surged during the first half of the year, increasing twofold compared with the last half of 2024.

In September, Citi added agentic tools to the Citi Stylus Workspaces platform, an in-house research assistant powered by Anthropic’s Claude model and Google Gemini. The bank gave its developers access to an AI coding assistant in January and its wealth division furnished staff with two AI-equipped advisory platforms last month.

Royal Bank of Canada credited its RBC Borealis research institute with securely training a financial services foundation model called Atom that is projected to contribute to the company’s goal of generating at least $700 million in value from AI by 2027, according to a July announcement. FICO rolled out a pair of AI models trained to assuage reliability, transparency and governance concerns, according to a September announcement.

The focus among the major players is on research that targets banking processes, Mousavizadeh noted.

“Through their research programs, banks like JPMorgan Chase, Capital One, RBC, Wells Fargo and TD Bank are setting the tone for how AI will be deployed in high-stakes, regulated environments,” Mousavizadeh said. “In contrast to the more commercially guarded [research and development] practices of Big Tech, these banks are signaling the future of applied AI in financial services and, most impressively, moving from research pipelines into production at scale within two to three years, which is lightning fast by academic standards.”

Information technology teams at financial companies have their hands full untangling years, and sometimes decades, of tech debt. Reliance on legacy technology can slow innovation and put broader business goals out of reach.

Artificial intelligence can step in to speed up the modernization process, said Roman Eisenberg, head of technology for Chase card and connected commerce at JPMorgan Chase. 

“The core of credit card processing served us well for many decades because it was built for scale and extendability,” Eisenberg said. “But it was built on mainframe, using legacy programming languages.”

Over time, however, the mainframe infrastructure made it harder to innovate. Chase was also grappling with skills gaps, making it difficult to maintain let alone expand capabilities.

While initially skeptical about including AI as part of platform modernization efforts, Eisenberg and his team were able to allay concerns through robust governance and guardrails. 

“At first, the criticality, the risk associated with making mistakes and all kinds of hallucinations that we saw, there was no way we could allow it to be anywhere near mission critical [platforms],” Eisenberg said. 

Other enterprise leaders can relate. AI adoption that isn’t supported with adequate safeguards and processes can degrade outcomes. Analysts have urged technology chiefs to approach the opportunity with care. 

The potential upsides motivate CIOs to keep trying.

AI-assisted modernization can help organizations unshackle themselves from outdated systems at a faster pace. Around 80% of IT and business leaders expect AI coding assistants to help document legacy applications, rewrite old code and automate software testing, according to a Publicis Sapient report published in May. 

Identifying governance gaps and establishing the right safeguards changed everything for Chase.

“We saw immediately that it allowed us to unlock real value for our customers,” Eisenberg said. “For me, it was one of those ‘a-ha’ moments.”

Standing up guardrails

The stakes are high for modernization projects, but success is hard to come by. 

The majority of legacy tech upgrades fail initially, and more than half contend with the stalling of broader digital transformation initiatives, according to a Forrester survey. Mainframe skills gaps, complex integration strategies and inadequate tools are all to blame for misfires. 

“We cannot make mistakes,” Eisenberg said. “You have to move fast, but you absolutely have to minimize any chance of breaking things because it’s how people buy groceries and pay for childcare.”

With that in mind, Chase ensures upgrades occur in a sandbox before moving into production. With tens of millions of lines of code to refactor, teams go after the highest-value modules first. 

“A lot of it pertains to customer experience, so we would prioritize that to narrow it down,” Eisenberg said. “We then create that test ecosystem and apply generative AI tools to explain what the code does.”

From there, teams use the documentation to generate new code in modern languages that are tested, compared and validated. 

“If we see that the same tests work, we have a very high degree of confidence that we didn’t miss anything and that the generative AI didn’t hallucinate,” Eisenberg said. 

Humans in the loop are also responsible for ensuring that newly generated code is up to par. 

“The new code might have to be refactored further or moved to a different domain,” Eisenberg said. “But these are several steps that decide whether it’s ready to be shipped into production.”

Chase card teams have been using the process for about a year and are likely to continue into the foreseeable future. 

“It will end when we decommission the last line of mainframe code,” Eisenberg said. “But we’re not going to stop at mainframe. We have so much other infrastructure where this approach is beneficial.”

Amid plenty of artificial intelligence buzz, establishing a robust foundation for responsible AI innovation is essential, said Rathi Murthy, the new chief technology officer at Varo.

Implementing new technologies only where they deliver genuine value to customers is crucial in Murthy’s eyes.

“I have a strong belief that technology is not built for the sake of technology. Technology has to have a purpose,” Murthy told Banking Dive. 

After joining Varo, Murthy saw the need for two critical roles within the company: a chief architect to ensure the technology architecture remains scalable and secure, and an engineering strategy and operations executive. She hired FS Nooruddin as Varo’s chief architect and Payal Carpenter as an engineering strategy and operations executive. Both previously worked with Murthy at American Express, Gap and Yahoo, and are helping her establish necessary diligence and structure at Varo.

The San Francisco-based digital bank has been using data and AI for its underwriting, fraud detection and risk compliance, while also incorporating personalization across the customer’s journey. 

Responsible AI makes the “right nudges [to] this huge opportunity…to make this experience about you and me,” rather than cookie-cutter, she said. This is an area that Murthy wants to explore and closely monitor.

Still, Varo executives don’t rush into new technologies unless they provide “meaningful impact” to its customers, she added.

Murthy took the helm of the technology unit at Varo in March. She replaces Sachin Shetty, the former tech chief who left the company in August 2024.

In April, Varo’s former CEO, Colin Walsh, handed the baton to Gavin Michael. 

Like Michael, Murthy emphasized the importance of AI and machine learning in underwriting for the neobank’s low-dollar loan offerings, fraud prevention and detection, and in delivering data-driven customer experiences with enhanced personalization. 

“We are building out a leadership team around Gavin that can help us go faster and smarter and build a team that is bold in their thinking and responsible risk-taking, and curiosity to take advantage of innovation where it matters,” which she referred to as “a change.” 

Leveraging experience

Murthy was CTO and president at Expedia for about three years before joining the bank. She’s also been CTO at Verizon Media and Gap, and chief information officer at American Express. 

At Varo, Murthy plans to leverage her experience at Expedia, which partnered with OpenAI as an early adopter of ChatGPT plugins. The team she headed at Expedia built models on top of existing systems to ensure security and governance and used AI to solve customer problems, like summarizing thousands of hotel reviews. 

“I'm always looking at smarter ways that we can take advantage of modular technology to build faster,” Murthy said. She wants to focus on building differentiators for Varo. 

“Data and AI is our secret sauce,” she said, adding that Varo’s charter adds to a winning formula.

Murthy said a focus is “protecting [the charter] and making sure we focus on going as fast to deliver value through our partners, through buying the right technology and building the areas that matter.”

Varo became the first nationally chartered all-digital consumer bank in July 2020. 

The company, built as a digital-first platform, has an edge over other banks trying to maintain legacy systems while balancing innovation and creating new features, Murthy asserted, saying the “trade-off ... is very challenging” for those banks. 

Varo’s cloud-native technology differentiates it from its competitors, while its modular architecture helps the bank to focus on “being purposeful” in what and how it’s building, Murthy said. 

The platform’s modular architecture helps to break down the banking software into smaller, independent components, while each component handles specific business functions and can be updated independently. 

Varo intends to lean on things like its partnership with Visa to offer “friction-free” services to its customers, Murthy said. Varo may add other services to support its core banking aim, such as lending products with faster decision-making, she added. 

Good customers vs. bad

Customer feedback is also important, and Murthy is keen to listen to customers to understand the patterns that draw more “good customers” and keep the bad ones at bay. 

“The better our models get, the better our data gets, the more you are able to take in more signals,” Murthy said. “You can determine fraudsters proactively and make it friction-free for the ones who are coming with the right intention.”

She wants to ensure Varo’s tech strategy prizes customer experience, addressing their needs through in-house innovation or partners whose services enhance customer experience.

Varo seeks to embed security protocols and automated compliance within its technology, using identity management and data protection tools while enhancing monitoring capabilities. The bank is working to mature its Supernova model, the in-house decision-making platform, to better interpret customer responses. 

“We just need to keep feeding the engine and maturing the engine as we go,” Murthy said.

Murthy has established an “architecture guild” at the company that reviews all new features, technologies, and tools; it includes participants from across product and engineering teams, not just leadership. The team is tasked with flagging any security concerns for additional testing.

“We are continuing to lean in on building out the products that we have, and continue to drive those products in an exceptionally well manner,” she said. 

Catena Labs, co-founded by Circle co-founder Sean Neville, recently raised $18 million to build an artificial intelligence-native bank designed to serve the emerging AI economy.

Catena has not yet released a product, but has published an open-source agent commerce kit that outlines protocols for “agentic commerce,” or AI-driven e-commerce. Catena plans to use the new funding to speed up the development of its AI-native financial institution and its underlying technology. 

During an interview, Neville, Catena Labs’ chief executive, said the startup is working toward its next major milestones: product releases and acquiring necessary financial licenses.

Catena aims to collaborate with and help AI-powered businesses operate safely and compliantly, pursuing becoming a licensed financial institution, Neville noted. 

Since he started Circle, the financial services industry's appetite for technology has changed. This is evident in the stablecoin bill that’s advanced in Congress and companies like Stripe embracing stablecoins. 

“A proper understanding of compliance and regulatory risk needs to have a seat at the table alongside product decisions and strategy and engineering and marketing and sales,” Neville told Banking Dive. “That's not necessarily true in other kinds of software development, but it's certainly true in fintech and building a viable financial solution. So we're bringing the same approach to Catena to what we're doing today.”

The agentic commerce kit released in May provides building blocks that help to identify, what is sometimes called “know your agent.” Catena plans to use the identity mechanism to ensure that when a business or individual encounters Catena’s AI agents — whether they're integrated into supply chain workflow or used by an e-commerce shop — the customers can verify that it is acting on behalf of the company or not. It also has protocols on how an automated system requests and executes payments. 

Neville envisions that payment activity will move away from traditional brick-and-mortar credit card models adapted for e-commerce and shift to a new model specifically designed for AI actors to execute payments and transactions.

Neville noted that Catena aims to play a pivotal role in this shift, adding that global exchange fees for transactions anywhere should be low, while safety should be enhanced.

Why now?

The time is right to build the an AI-native bank, since we are at the “nexus of a new kind of web that is very AI-driven,” Neville said. 

He said he believes there will be a time in the near future when AI agents conduct most economic transactions, requiring a completely new financial infrastructure designed from the ground up to accommodate this reality. 

The web is going to be agent-native, he expects, so it’s the right time to build identity systems, an understanding of risk, proper guardrails, and oversight surrounding compliance and letting AI actors transact with humans and one another.

Challenges of AI-native financial institutions

As AI use evolves and grows, Neville said he believes banks will have a system of AI actors performing certain tasks.

Catena plans to use AI actors, or digital workers, as employees to perform internal tasks like writing software, and humans for oversight and escalations, such as risk monitoring, compliance management, anti-money laundering, and filing suspicious activity reports, Neville said.

The AI-native bank is designed to serve AI businesses that execute transactions. However, certain limitations prevent AI usage and need to be addressed: AI actors can't get fingerprinted to become licensed money transmitters, for example, and most risk infrastructure is explicitly designed to prevent bots from using financial systems, he said.

Neville said he thinks the system should be built so that AI actors will be the primary users, instead of blocking them. Some key challenges need to be tackled, like properly identifying AI actors, mapping AI actors to real-world legal entities, including businesses and individuals, maintaining security by keeping out malicious actors, and creating authentication and authorization systems for AI, he said.

Catena’s financing round was led by a16z crypto, with participation from Breyer Capital, Circle Ventures, Coinbase Ventures, CoinFund, Pillar VC, and Stanford Engineering VF. The funding round was also backed by angel investors, including Tom Brady, Bradley Horowitz, Hamel Husain, Kevin Lin, Peter Mattoon, Sam Palmisano, and Balaji Srinivasan.

“Integrating AI into commerce demands a rebuild of financial infrastructure,” Jim Breyer, founder and CEO of Breyer Capital, said in a statement in May. “Legacy systems cannot always keep up — and that’s why Catena is so compelling.”