Skip to content
Banking & Compliance

Agentic AI Meets The Bank Login: Schwab, Citi, Wells Fargo And Canada's Big Four Draw Lines On Customers' AI Agents

In three weeks, Schwab, Citi and JPMorgan have blocked outside AI agents from moving customers' money, Wells Fargo has written the risk onto customers from 24 November, and RBC, TD, CIBC and Scotiabank have said they will block or limit agents too. A side-by-side view of what each bank has done, why, and what it means for firms building agentic AI for finance.

AlchmAI Editorial7 min read

An illuminated yellow Bank sign on the side of a building at night

10 to 20

Times a week Schwab says it blocks unauthorised agents trying to move client money (Hunton, 3 November 2026)

24 Nov

Date Wells Fargo's new terms treat agent-initiated transactions as authorised by the customer (Hunton)

4

Canadian banks, RBC, TD, CIBC and Scotiabank, that say they will block or limit third-party AI agents (MobileSyrup, citing CBC, 2 November 2026)

1 Jan 2027

Start of JPMorgan's updated agreement making customers responsible for what their AI agent does (Hunton)

Agentic AI has reached the bank login, and banks are answering in two ways. Between 13 and 20 October, JPMorgan Chase, Charles Schwab and Citigroup moved to stop outside AI agents acting inside customer accounts, according to a 3 November client alert from law firm Hunton Andrews Kurth. Wells Fargo chose a contract instead of a block: from 24 November its customers use agents "at your own risk". On 2 November MobileSyrup, citing CBC News, reported that RBC, TD, CIBC and Scotiabank will block or limit third-party agents too.

We covered JPMorgan's first move on 14 October. Three weeks later the picture is wider and clearer: most large banks will allow agents to read, few will let them move money, and the ones that allow more are writing the liability onto the customer.

What Have Banks Done About Customers' AI Agents?

Most have kept agents to read-only access and kept money movement for themselves. The table sets out each bank's position as reported.

How large banks are treating customer-directed AI agents
BankPositionCustomer liabilityDate
JPMorgan ChaseOutside agents not allowed to act inside accounts; read-only data access through providers such as Plaid remainsUpdated agreement: a customer who gives an agent access authorises everything it doesAgreement from 1 January 2027
Charles SchwabBlocks unauthorised agents that try to move money, 10 to 20 times a week; building controlled connections without money movementNot described16 October 2026
CitigroupDoes not permit third-party agents to log in or transact; limited read-only access via open banking providersNot described16 October 2026
Wells FargoNo block; may limit or end agent access at any time; no duty to monitor agentsAgent transactions treated as authorised; use is at the customer's own risk24 November 2026
RBC, TD, CIBC, ScotiabankWill block or limit third-party AI agents on online bankingNot describedReported 2 November 2026
Source: Hunton Andrews Kurth client alert (3 November 2026); MobileSyrup citing CBC News (2 November 2026)

The language matters. JPMorgan chief executive Jamie Dimon cited "rogue agents" on the bank's third-quarter call. Citi said it does not permit "third-party AI agents or assistants to log into a customer's account". Wells Fargo's new Online Access Agreement has a section titled "Third-Party Artificial Intelligence Agents or Agentic AI", which says agent use is "at your own risk".

The detail behind each position differs. Dimon named Meta's Muse and Instinct as agents that will not be allowed to act inside customer accounts, and a JPMorgan spokeswoman said the bank does not let third parties obtain customer credentials and log in on their behalf. Schwab chief executive Rick Wurster told Bloomberg the firm is developing controlled connections that give agents information without money movement, and Schwab has launched its own assistant, Charley. Wells Fargo's terms also disclaim any duty to monitor agent activity and reserve the right to limit or end agent access at any time.

Why Are Banks Blocking Agentic AI Now?

Because agents arrived before the rules did. MobileSyrup, citing a Reuters report of 22 October, says some US banks, including Morgan Stanley and Wells Fargo, found unauthorised AI agent activity on customer accounts, with some agents allegedly acting without user instructions. It also cites a severe flaw reported in Meta's Muse agent that could let an attacker take over the AI and act on a victim's device. Customers had been able to connect agent tools to bank accounts and credit cards, although purchases usually still required a person to approve them. RBC chief executive Dave McKay wrote that autonomous systems raise risks for fraud, consumer harm and the wider economy.

The regulatory gap is real. Hunton reports that bank executives raised the growth of agent-initiated transactions with the Federal Reserve, and that regulators privately welcomed the restrictions while indicating that customer authorisation and liability may need legal clarification. The US open banking rule that would have set data-sharing terms was enjoined by a federal court; its 1 April 2026 compliance date passed without effect, and a reconsidered rule went to the White House budget office in August 2026 without being published.

How Does This Compare With Brokers And The UK?

Brokers are moving the other way. Interactive Brokers, Robinhood and others now let connected AI platforms trade in client accounts, in some cases without approving each order, as we reported on 27 October. The difference is structure: a brokerage can ring-fence an agent in a separate account with its own limits, while a bank login usually opens everything.

The same split runs through every firm that holds customer money: reading is low risk and widely allowed, acting is high risk and tightly held. What changes the answer is not the model behind the agent but the plumbing around it: separate credentials instead of shared passwords, permissions that can be narrowed to one action, limits enforced by the firm rather than the agent, and a log that shows who instructed what. Where that plumbing exists, firms have been willing to go further.

In the UK, no bank has announced a comparable policy in the sources we reviewed, but the FCA has said the issue is coming. In his 22 October Mansion House speech, chief executive Nikhil Rathi said personal AI agents are changing how consumers use financial services and are expected to reach UK consumers soon. The FCA's stated approach is to apply existing rules, so a UK firm's terms, authentication and complaints process are where it will be tested.

Our View: Read-Only Is The Easy Answer, And It Will Not Last

Blocking money movement is a sensible first response to agents that log in with a customer's password. It is not a long-term position, because customers will keep using agents and competitors will offer safer ways in. The durable answer is the one Schwab hints at: controlled connections where the bank, not the agent, defines what is allowed, with limits, revocation and a record of every action. Wells Fargo's approach shows the alternative, which is to allow access and move the risk onto the customer, and that is likely to draw attention from consumer regulators.

For firms building agentic AI in finance, the lesson is to assume the bank or broker will ask three questions: how does the agent authenticate without the customer's password, what exactly can it do, and how is every action logged. The products that answer those first will get access first.

“Banks are not banning agents; they are banning agents that log in as the customer. The firms that offer a safer door will be the ones let in.”


What It Means For Trading And Fintech Firms

  • Brokers and CFD providers: if you allow agents, give them their own credentials, scopes and limits, never the client's login. That is the difference between the brokers opening up and the banks closing down.
  • Prop trading firms: funding and payout flows run through banks and payment providers that are tightening agent access. Check that any automated payout or funding step does not depend on an agent logging in as a person.
  • Compliance leads: review your terms for agent use. Decide whether you, like Wells Fargo, treat agent instructions as the customer's, and make sure customers can understand it.
  • Fintech and trading startups: build for delegated, revocable permissions and full action logs. Read-only access is available now; money movement will follow only where banks can see and stop what an agent does.
  • Research teams: agent traffic is becoming visible in bank and broker data. Separating agent and human activity is now a reporting requirement in waiting.

The Bottom Line

Agentic AI has met the bank login and, for now, the login is winning: JPMorgan, Schwab and Citi block outside agents from moving money, Canada's four largest banks plan to block or limit them, and Wells Fargo allows access from 24 November at the customer's own risk. AlchmAI, a London trading technology and AI studio, builds agent access with scoped permissions, limits and audit logs for brokers and fintechs, and the banks' message this month is that the firms with those controls will be the ones trusted with more than read-only access.

References & Further Reading

Banking & ComplianceAgentic AIAI agentsopen bankingcustomer authorisationAgentic AI London
Share Email
AI

AlchmAI Editorial

Research and analysis, London

The AlchmAI team writes about the markets, technology and regulation we work with every day. We build trading platforms, real-time charts and AI analysis tools for brokers, prop firms and fintech teams from our office in Mayfair, London. Every article lists its sources. Nothing we publish is investment advice.

This article is general information and commentary. It is not investment advice or a recommendation to buy or sell any investment. Important information