The alarm began as major banks detected third-party AI agents entering customer accounts, often after customers shared their own credentials. Executives told Axios they could not always tell whether an agent was carrying out a customer’s request, had exceeded it, or had gained access improperly — a gap that has pushed Morgan Stanley, JPMorgan, Bank of America, Citi and others to restrict the tools.¹

The concern is less about a confirmed theft than about what banks cannot yet see. A bank executive said institutions struggle to distinguish among agents and determine whether one was authorized or had “gone rogue,” while a separate account of the industry’s response described the resulting uncertainty as a major operational headache.² JPMorgan has urged customers not to hand passwords to “any person or thing,” even as banks explore read-only connections that would let services view balances without moving money.³

That defensive posture has drawn a sharp distinction from David Sacks, who said there is “a world of difference” between user-authorized agents and genuinely rogue systems that could amount to cyberattacks. In his view, much of the activity described falls into the former category — meaning banks risk treating customer choice as a breach.David Sacks (DavidSacks)⁴

Wall Street’s wider fear is that agents could make it effortless for customers to hunt higher yields and shift deposits, squeezing bank profits. But the evidence remains thin. RBC’s Gerard Cassidy calls unrestricted money-moving access “too risky,” while Wells Fargo analyst Mike Mayo says the threat is serious but “exaggerated”; third-quarter results showed no clear deposit disruption.⁵

For now, the divide is over sequencing: banks want safeguards and clear liability before automation gets transaction power. Their critics want a path for customers to grant, limit and revoke that power — rather than a blanket digital lockout.