Jackie Reses, co-founder and CEO of Lead Bank, told CNBC that “There’s no actual conversation of how AI needs to change in order to execute banking in the future.” Regulators have published material on artificial intelligence, but most discussion has stayed at the level of principles rather than concrete revisions of specific rules.
What Lead Bank Put on Paper
Reses described her firm’s response: “We took pen to paper and actually made proposals for every place we believe laws and regulations should change, so that we could start the conversation in the banking industry about what has to change when agents are driving transactions instead of humans.”
First, Reses runs a privately held bank and proposes rules that would govern banks like hers (ordinary advocacy with a commercial stake). Second, this is one company’s proposal. No regulator has adopted it, and no legislature is considering it.
How AI Agents Handle Your Money Today
Jackie Reses explained the current setup: “Today it happens, but it’s happening at a transaction level. You can execute a transaction if you ask an agent to do a buy or sell… It does not have unfettered control over accounts, and it doesn’t have any access to accounts unless you specifically give it access.”
Reses describes a narrow model of permission for a single task, distinct from standing authority over a checking or brokerage account. You tell the agent to buy or sell, it carries out that one instruction, and its access stops there unless you grant more. That changes the risk. A one-task permission limits an error to one transaction. Standing authority puts the whole account at risk.
Why Less Deterministic Transactions Raise New Regulatory Questions
Jackie Reses then described where things are heading: “We are well on our path to having transactions that are less deterministic. And so we need to develop laws and regulations in order to enable that to happen.”
A deterministic transaction follows a rule with a predictable outcome: pay a bill on a set date, or sell if a price falls below a set level. A less deterministic transaction lets the agent use judgment, such as deciding which bills to pay first when cash runs short, or when to rebalance a portfolio, according to Lead Bank.
That difference is the core regulatory problem. Financial rules assume a human or rule-based system is responsible for each decision. Authorization, error resolution, and fraud liability trace back to a person who approved the action. Once software picks the action, new questions arise: who authorized it, who bears losses if it goes wrong, and what counts as consent. Today’s one-task model avoids these questions because a human still makes each call.
Can Banking Become AI’s Governance Template?
Jackie Reses argued that “the finance industry can be an incredible template for what governance should look like in AI.”
Banking already runs on mandatory controls, audit trails, identity requirements, and accountability for errors. Most sectors using AI have far less in place. However, banking rules change slowly, differ across countries, and were written for institutions with software as an afterthought. The Consumer Financial Protection Bureau’s draft strategic plan commits to studying “emerging market areas of alternative products or services with high growth rates and potential risk to consumers”, a broad mandate with no rules specific to agents. Banking offers the best existing model for controls, but revisions accountability rules built around human decision-makers will be the hard work.
Useful Contribution or Rule-Shaping Bid? What to Watch Next
A concrete draft gives regulators and peers something specific to argue over, advancing a discussion Reses sees as stuck. It also reflects the priorities of the bank that wrote it.
Three signals will show whether the proposals gain traction:
- Whether any regulator publicly acts on with the proposals.
- Whether other banks publish competing frameworks.
- Whether agent authority grows from one-task permission to standing account access.