Banks Race to Deploy AI While Managing Mounting Risks
Banks worldwide face a critical choice: move fast on artificial intelligence and risk costly failures, or move cautiously and fall behind competitors. The stakes are unusually high for an industry that underpins the global economy.
More than 90% of banks have already started AI projects, according to Sameer Gupta, Americas financial-services AI leader at EY. Yet nearly half of those initiatives fail to deliver returns, says Miriam Fernandez, S&P Global's lead researcher on AI adoption in banking.
The potential payoff justifies the effort. S&P Global estimates that banks' average return on equity could rise to 14% from 12% over the next three to five years. But getting there requires overhauling how large, data-heavy organizations work-a task that is both expensive and slow.
The Internal Efficiency Play
Banks are starting with internal operations, where risks are easier to contain. More than 85% of current AI use cases stay inside the organization, according to Alexandra Mousavizadeh, co-founder and co-CEO of Evident Insights, a London consultancy tracking AI in banking.
An AI agent might cut client onboarding from six months to six weeks by automatically retrieving and cross-checking documents for know-your-customer compliance. Investment bankers could save hours gathering analytical and legal documentation for mergers. Treasury and custody operations could run more efficiently.
These gains won't make headlines outside finance. But they directly affect the bottom line. "We'll see efficiencies first before we see a lot of visible innovation," says JoAnn Stonier, former chief data officer at Mastercard and now professor at Carnegie Mellon University.
Security Risks Multiply When Systems Connect
The real danger emerges when banks move beyond their own operations. Customers will eventually want their own AI tools to interact directly with banks' systems. Payment networks connecting hundreds of banks and millions of merchants create even thornier problems.
"Agents have to work across an ecosystem where Mastercard is using one large language model and the counterparty another," Stonier says. "We don't have the protocols yet for how they metaphorically shake hands."
In April, US Treasury Secretary Scott Bessent and Federal Reserve Chairman Jerome Powell summoned top US bankers to discuss cybersecurity risks tied to Mythos, an AI model from Anthropic. The model had uncovered flaws in computer operating systems that survived decades of human review and millions of automated security tests. Anthropic agreed to restrict access to a small group of corporate clients for defensive purposes only.
Rogue AI agents could infect or co-opt other systems. "Risk is coming in through the back door, with vendors' agents liaising with each other," Mousavizadeh says. Firewalls may not be sufficient to stop it.
The Speed Mismatch
Banks and technology companies operate on different timelines. Financial institutions plan three to five years ahead. The tech sector moves in six to 12 month cycles. "By the time a tool is deployed, it's obsolete," says Eric Alter, who recently retired as an AI engagement leader at Marsh.
Regulators lag even further behind. They typically react forcefully only after a crisis-which could be too late given AI's potential power. "It's very difficult for the laws to keep up," Stonier says. "It's up to organizations to retain the trust of customers so they can stay in business."
Data and Talent Determine Winners
Banks that move fast enough to capture AI's benefits could build lasting competitive advantages over slower rivals, according to S&P Global. The agency expects financial and competitive positions to diverge significantly within three to five years.
Two factors separate leaders from laggards: data readiness and access to AI talent. "Data readiness, working with data sets that are clean and not duplicated, is a source of competitive advantage now," Fernandez says. AI models are only as good as the data they consume.
US banks have a natural advantage in hiring. Seven of the top 10 banks in Evident's latest AI Banking Index are headquartered in North America, led by JPMorgan Chase, Capital One, and Royal Bank of Canada. "European banks are one step behind, without the same access to an AI startup ecosystem," Mousavizadeh says.
The talent concentration is striking: just three US banks-JPMorgan Chase, Capital One, and Bank of America-account for 75% of the industry's AI-related patents.
Customer-Facing Wins Emerging in Europe
Some European banks are compensating for smaller talent pools by focusing on customer-visible applications. Germany's Commerzbank deployed an AI customer service avatar named Ava a year ago. It handles more than 30,000 inquiries monthly and resolves three-quarters of them.
Italy's UniCredit built DealSync, a platform that identifies merger and acquisition opportunities for midsize companies across Austria, Germany, and Italy. Dutch bank ING uses AI to smooth banking call centers, routing sensitive life events to human staff while handling routine questions automatically.
ING plans to enable mortgage applications entirely through AI agents this year, according to Chief Operations Officer Marnix van Stiphout.
Trust May Matter More Than Speed
Internet adoption took roughly a decade before banks and other institutions fully integrated the technology. Mousavizadeh expects AI to diffuse about twice as fast-still five years of gradual rollout. "It's never a plug-and-play process," she says.
One lesson from the internet era: banks proved more durable than the tech companies that disrupted them. Most global tech giants barely existed 25 years ago. The largest banks then are largely the largest banks now.
That stability stems partly from trust. Societies cannot function without banking systems. In an era of widespread distrust in government, media, and technology, that institutional trust could be a more valuable asset than speed. One major failure with AI could damage a bank's franchise permanently.
Proceeding carefully makes sense. "Moving so fast, it's very hard to get the right balance of risk and innovation," Fernandez says. But standing still is not an option either. Banks must adjust their operations and cultures to absorb AI gradually, learning from early mistakes before they become catastrophic.
For finance professionals, the takeaway is clear: AI adoption in banking will be neither quick nor painless, but it will be relentless. Those who understand both the efficiency gains and the security risks will shape their institutions' competitive positions for years to come.
Learn more: AI for Finance and AI Agents & Automation cover practical applications in financial services.
Your membership also unlocks:
