Even as most financial institutions are still trying to figure out whether they want to use artificial intelligence agents, it has suddenly become clear they will have no choice but to deal with their customers’ personal AI agents. That raises a host of risk and competitive concerns for banks.
In just the past three months, several major technology providers have released platforms that offer personal AI assistants to consumers. Those include SpaceXAI’s Grok Bot, Meta Platforms’ Muse, OpenAI’s dots and Anthropic’s Claude Duet. Within just 1.5 months of launch, more than 10 million people downloaded Muse, according to Meta’s Oct. 21 earnings call.
Larry Pruss, managing director of emerging payment technologies for the strategic advisory firm SRM, says he is already using Muse to book Uber rides and airline flights. “My agent has a credit card through Muse, and it is using that credit card to make purchases on my behalf,” he says.
Rajita Iyer, executive vice president for North American services at Mastercard, said at Jack Henry & Associates’ Connect 2026 last week that such consumer behavior is already having an impact. “On the Mastercard network, we saw a six times increase in agentic commerce transactions just in the last six weeks,” she said. “When that trust gets established — and it could happen very quickly — that things will start scaling very quickly.”
Ben Metz, Jack Henry’s chief technology officer, told the audience at his mainstage presentation that such scale is inevitable. “This is coming like an army to your online banking systems; like an army,” Metz warned.
Since Jack Henry provides digital banking platforms for more than 1,000 banks and credit unions, Metz said he is keenly aware of the company’s responsibility to help those institutions get ready for the invasion. “And we’ve been working over the last year and a half to prepare for this. We thought it would happen — I honestly didn’t think it would happen this fast.”
Banks Aren’t Prepared
A recent study from Encore AI, a company that builds AI agents for financial institutions, found that most banks are not ready to handle consumer AI agents. Encore researchers sent personal AI agents to the websites of more than 1,200 organizations to find out what they could and could not accomplish. Banks ranked 18th out of 23 industries in terms of AI-agent readiness based on parameters that included discoverability, access and web experience.
Encore AI Founder and CEO Dvir Ginzburg says 25% of the banks they tested actually block AI agents at the gate. That means the agents aren’t allowed to enter through connection portals traditionally accessed by data aggregators and screen scrapers. “For the past 20 years, we’ve been building software that was blocking bots,” he says. “But now, bots are actually clients in disguise.”
Ginzburg says banks, though, are having a tough time blocking agents. “We see customers asking their personal assistant to open a CD, to open a deposit, to ask for a pre-approval for an auto loan. So, the clients are doing that, and because Muse and Grok Bot are so capable, [they] are actually finding ways around these blocks,” he says. “A personal assistant can actually open up a browser and act as a human, and that’s something that is much harder to block.”
An AI assistant can then potentially log in to an account if they have the proper credentials, which Ginzburg says some people are providing to their agents.
Erica Pilon, vice president and head of corporate strategy for Jack Henry, uses Instinct, an invite-only, text-based personal AI assistant. But she will not provide it with her banking credentials at this early stage. Pilon advises others against doing so as well. “I’m not comfortable with that yet, but I think that will move very fast,” she says. “If you look at the analogy of e-commerce. . . when that first came out, people were like, ‘I’m not putting my credit card information online because I don’t know where it’s going, and I don’t want to pay for something this way.’ Now, I mean, would you ever think of not [doing it]?”
Pruss believes that as more consumers use agents, financial institutions will have to ensure that websites are built so bots can include them in comparison shopping tools and put the bank in a good light. That is why Ginzburg advises banks and credit unions to not only allow personal AI agents in but to also implement agent protocols issued by the various technology providers. Those make it easy for personal assistants to navigate an institution’s website or digital banking platform.
“You can tell the agents, ‘For rates, you should go here. For pre-approvals, you should go there,” Ginzburg says. “Banks that are going to allow these maps and instructions are actually going to get more traffic.”
Security and Liability
Allowing personal AI agents with easy access may boost visibility and competitiveness, but it currently presents unknown risks as well.
“They’re not malicious, but we somehow need to work that out because some [bots] are,” Ginzburg says. “When we are starting to work with a new client, with a new bank, they don’t even know how many requests from personal assistants they got because everything is blocked the same way.”
When it comes to liability, Pruss says he assumes he’s on the hook if Muse makes a purchase he didn’t want to make. J.P. Morgan Chase & Co. Chairman and CEO Jamie Dimon indicated that is the correct assumption on the bank’s October third-quarter earnings call. “The people [who] are using them, they should understand the risk,” he said. “There’s a liability transfer; they should understand the liability transfer.”
While it’s not clear all consumers understand the potential liability involved, a survey out this week from the American Bankers Association shows 77% of consumers are concerned about giving an AI tool or agent permission to move money, make payments or carry out other financial transactions for them. But 27% already say they are likely to give an AI agent that capability.
Mastercard’s Iyer said there are also concerns over how to verify both identity and intent in agent transactions. “Do we know who is behind that transaction or that interaction? Do we have an understanding of what was the intent of the human, and what parameters the human gave that agent to act on their behalf?” she asked. “That’s really important. Otherwise, the financial institution is completely blind to what’s happening.”
Metz said the solutions Jack Henry is developing within its existing systems will help community banks and credit unions handle those very issues. Jack Henry Identity is being upgraded to authenticate AI agents, while another service will help institutions determine an agent’s permissions.
“If you ran a workflow and there was a problem, or the regulator wanted to go in and look at it, what are they going to say? They’re going to want to see the human that was responsible for creating that workflow and the human that was responsible for running it and approving it,” Metz said. “Agents are going to be no different, and we had to build a control layer.”
Ginzburg says every small financial institution will need that type of assistance from its digital banking provider. “Yeah, 100%,” he says. “This is not something that they can do themselves.”
Previous