By Lynn Räbsamen, CFA | Advisory Board Member, CFA Institute | Author, Artificial Stupelligence
Within 7 days, 3 of the largest US banks clarified which AI agents get through the front door. Then the UK regulator published the house rules for what happens inside. Neither was dramatic. Both were useful.
In short: In mid-October 2026, JPMorgan, Bank of America and Citi set limits on outside AI agents acting inside customer accounts. In the same week, the UK’s Financial Conduct Authority published good practice for governing AI. The message is consistent. AI can act, but a named human stays accountable, especially when money moves.
For wealth firms, that means clear ownership, meaningful review and vendor oversight, starting now.
One Week, Two Signals on AI Agents
Some weeks, the future arrives quietly. This one arrived on a JPMorgan earnings call.
The backdrop is simple. AI agents no longer just answer questions. As the FCA’s Jessica Rusu put it in a speech on October 19, “Agents can search across products, manage payments and investments, and transact directly with other agents.”
Software that can autonomously move money raises an old question in new clothes. Who answers for it?
The Front Door: What the Banks Said About AI Agents
JPMorgan: No Rogue Agents
On JPMorgan’s third-quarter earnings call on October 13, Jamie Dimon was characteristically direct. “We’re not going to allow just any outside agent to come inside the company,” he said. “You don’t want rogue agents.”
A spokesperson supplied the operational detail. When the bank detects a third-party agent trying to access accounts by obtaining customers’ banking credentials, “we do not allow it.”
“You don’t want rogue agents.” Few risk frameworks have been summarized so efficiently.
Bank of America: A Human in the Loop
Chief executive Brian Moynihan told CNBC: “We strongly believe there should be a human in the loop to authorize money moving out of your account.”
The Financial Times reported, citing a person familiar with the matter, that the bank is limiting AI agents’ access to consumer deposits. Coverage of the move ties it to a question nobody has settled yet: who is liable for a payment an agent authorized?
Citi: A Policy Already on the Books
Citi did not announce anything new. It pointed to its existing policy. Third-party AI agents or assistants may not log into a customer’s account or “transact directly within our digital banking platforms.”
What This Is Not
This is not a ban on AI. JPMorgan said it supports “secure, customer-authorized data sharing through open banking providers.” Coverage of the restrictions also notes that customers can still grant limited, read-only access to account information through open banking providers.
Each bank spoke for itself, in its own words. The line they drew is still remarkably similar. Reading data with the customer’s permission is fine. Logging in as the customer and moving money is not.
The House Rules: What the FCA Expects From AI Governance
On 19 October, the FCA published “AI in financial services: good practice and areas for improvement.” It is not a rulebook. The FCA states that its examples “do not create new requirements.” Instead, it shows how existing rules, including the Consumer Duty and the Senior Managers and Certification Regime (SM&CR), already apply to AI.
The document is written for UK firms. Its logic travels well beyond them. Four points stand out.
Ownership Starts at the Design Stage
In the good-practice examples, senior managers “were accountable and owned risks from the design stage onwards for each AI use case.” One smaller firm “mapped every material AI use case to a named senior manager.”
Not a committee. Not a working group. A name.
Human Review Must Mean Something
The FCA flags a risk of human review of AI outputs “being a ‘box-ticking’ exercise, rather than meaningful oversight.” It adds that effective oversight is not simply about reviewing outputs. Reviewers need authority, information and a route to escalate.
A reviewer who approves everything is not a control. It is a delay.
You Can Outsource the Tool, Not the Responsibility
“Firms can outsource AI services, but remain responsible for them.” Buying a model does not buy you out of accountability. Put plainly: the vendor’s terms and conditions will not appear before the regulator on your behalf.
Proportionate Is Allowed
“Smaller firms can govern AI effectively with targeted, proportionate controls.” That is welcome news for boutiques, independent advisors and external asset managers. Good governance does not require a fifty-person AI office. It requires clarity about who owns what.
The Common Thread: AI Can Act, a Human Answers for It
Put the two signals side by side and one principle emerges, in three parts.
- Accountability sits with a person. Not with a model, a vendor or an org chart box labeled “innovation.”
- Human control matters most when money moves. Banks drew their line at transactions, not at data. The FCA draws its line at outcomes.
- Responsibility cannot be outsourced. You can delegate the work. You cannot delegate the answer to “who signed this off?”
None of this is new. Fiduciary duty predates the microchip. What is new is the speed at which software can act on a client’s behalf, and the temptation to let governance catch up later.
So What for Wealth Firms? Three Questions to Ask Now
1. Who is the named owner of each AI tool we use? If the answer is “IT” or “the vendor,” it is not an answer. Each use case needs a senior person who understands it and can explain it.
2. Would our reviewers catch a wrong output? Review only works if it can say no. Ask how often reviewers reject or change an AI output. If the honest answer is “never,” that is information.
3. What happens if our AI vendor changes or fails? The FCA treats AI dependencies as part of operational resilience. For firms in scope of SYSC 15A, mapping them is already a requirement. Know what the tool touches, who else depends on it, and what you do on the day it stops.
The Agentic Era Arrives With Guardrails
The agents are coming. They are simply being asked to knock first.
Firms that build accountability in now will treat this as a design choice. Firms that wait will treat it as a retrofit. Retrofits are rarely cheaper, and never more elegant.
FAQ
Did US Banks Ban AI Agents?
No. JPMorgan, Bank of America and Citi limited outside AI agents logging into customer accounts or moving money. Customer-authorized data sharing through open banking providers continues.
Does the FCA’s AI Good Practice Document Create New Rules?
No. The FCA says its examples do not create new requirements. They show how existing rules, such as the Consumer Duty and SM&CR, apply to AI.
Who Is Responsible When a Firm Uses a Third-Party AI Tool?
The firm. The FCA states that firms can outsource AI services but remain responsible for them, including their outputs and outcomes.
Can Smaller Wealth Firms Govern AI Effectively?
Yes. The FCA says smaller firms can govern AI effectively with targeted, proportionate controls, such as mapping each material AI use case to a named senior manager.
This article was partially drafted with AI, and reviewed and edited by a human.
For more insights about what AI can or cannot do, check out my book “Artificial Stupelligence: The Hilarious Truth About AI”.
Subscribe here to be the first to receive my insights.
