Agents that move money turn deposit stickiness into an assumption, not a fact

American Banker reports that technology companies including Robinhood and OpenAI are betting on AI agents to read consumers' financial lives, and in some cases to self-execute payments and rebalance investment portfolios ^[1]. The publication frames what is at stake for banks as disintermediation and deposit liquidity risk, not a customer-experience upgrade ^[2].

The distinction is the whole story. Daniela Hawkins, a partner at Capco, described the exposure using her own savings account: a super regional bank has never had to worry about her moving it because she is too busy working to think about it, "but she has [an AI agent that will do it for her], and then you multiply that over a million people, suddenly you've got material risk" ^[3]. Read that closely and the thing being priced in a retail funding franchise is customer inattention, not loyalty and not rate ^[1]. Software does not get busy at work.

The mechanism Hawkins describes is mundane: agents sweeping certificate of deposit and high-yield savings offers and moving money automatically, which American Banker notes could threaten lending operations at the bank ^[4]. That is not a channel problem. It is a change in the observed behaviour that sits underneath every non-maturity deposit assumption a treasury team carries, and the useful question for an operator is which of those assumptions are still measured and which are simply inherited from a decade when nobody rebalanced their savings on a Tuesday afternoon.

The gate today is permissioning, not appetite. OpenAI has said ChatGPT cannot make changes to users' accounts ^[5], while Hawkins' expectation is that "the next iteration of this is the agent that says, 'Would you like me to do this for you?'" ^[6]. On OpenAI's own account of demand, more than 200 million people come to ChatGPT every month for budgeting, investment questions, comparing paths and planning goals ^[7]. So the constraint on agent-led deposit movement is a product decision by a small number of vendors, not a shortage of people willing to ask ^[2].

Not everyone in the piece buys the endgame. Oban McTavish, co-founder and CEO of the transaction data enrichment firm Spade, said he does not believe banks will largely disappear as a service with everyone getting banking through a chatbot partnership with OpenAI or Anthropic ^[8]. American Banker also reports that while most banks are watching how customers use AI, there is little concern that agents will fully take the reins of consumer finances ^[9].

The nearer-term worry from an actual deposit-taker is different again. At New York-based Municipal Credit Union, Stuart Salembier, senior vice president of member engagement and channel delivery, said the institution's concern is members handing login credentials to large language models rather than agent-led deposit flight ^[10]. He said he advises members not to give financial information to a complete stranger and takes the same view of an AI bot ^[11], and that putting that information into an LLM carries significant potential for compromise, and with it identity compromise, with no clarity on where the data goes ^[12]. He also questioned whether ordinary consumers can prompt these systems effectively ^[13].

Watch three things. Whether OpenAI or a peer relaxes the no-account-changes limit ^[5], because that is the switch. Whether any bank starts disclosing an agent-driven behavioural assumption in its funding model rather than treating this as a roadmap item ^[2]. And whether credential-sharing losses arrive first, which is where Municipal Credit Union expects the damage ^[10].