Citigroup and Bank of America now restrict AI agents from customer bank accounts, joining JPMorgan Chase and Charles Schwab, according to reports tied to the Financial Times.
If you sell $2,000 to $10,000 products, this reaches your store in two places: how buyers pay when an assistant like Meta’s Muse shops for them, and how you let AI tools touch your own business bank account. I track both at Ecommerce Paradise because high-ticket orders are exactly where a payment rail failing, or a liability gap, costs real money.
The banks are blocking agents that log in with a customer’s password and move money. They are not blocking Shopify checkout. That distinction decides what you need to change this week, and what you can ignore.
Form your LLC with Northwest before you hand any AI tool a business bank login. Northwest Registered Agent is the formation service I point ecommerce owners to, and unlike cheap formation sites that upsell you, it forwards physical mail to operators who live outside the formation state or country. Form your LLC with Northwest →
Citi, Bank of America, JPMorgan and Schwab Limit AI Agents
Citigroup does not allow third-party AI agents or assistants to log into a customer’s account or transact through its digital banking platforms, per the Financial Times report republished by Yahoo Finance on Oct. 19. Citi says the policy protects customer accounts and information. Customers can still grant read-only access through certain open banking providers, according to The Paypers.
Bank of America is limiting agent access to consumer deposits while it builds protocols for who is liable when an AI authorizes a payment. Chief executive Brian Moynihan told CNBC: “We strongly believe there should be a human in the loop to authorise money moving out of your account.” The bank is also working with agent providers so consumers can trace money an agent moves.
JPMorgan Chase said on its third-quarter earnings call that it will keep outside agents out of client accounts until its own protections are ready. Chief executive Jamie Dimon said, “You don’t want rogue agents and there could be thousands of them,” per Banking Exchange. JPMorgan still supports customer-authorized data sharing through open banking providers.
Schwab got there first. Chief executive Rick Wurster told Bloomberg Television on Oct. 16 that the firm blocks unauthorized outside agents trying to move client money, including Meta’s Muse. His objection was liability: “We can’t control it, the client can’t control it, something could go wrong and who’s liable?” Schwab sees large agents blocked “in the tens and twenties” each week, per The Next Web. It plans read-only access through model context protocol servers, so outside tools can check balances but cannot send money.
The money behind the move is large. Apollo chief economist Torsten Slok warned that AI assistants could sweep household cash into accounts paying 3.3% to 5.0%, against a US average checking rate of 0.1%. JPMorgan, Bank of America and Wells Fargo hold $1.6 trillion in non-interest-bearing deposits, about 16% of their combined liabilities. The FT estimated that paying 3% on those balances would cost the three banks about $47 billion a year, nearly half their combined annual earnings, per OnInvest’s summary. That figure is a forecast from one analysis, not observed outflows.
Regulators are watching quietly. Wall Street executives have contacted Federal Reserve officials about agents handling financial transactions, and the Fed declined to comment, according to The Paypers. Regulators privately support the restrictions and believe the law may need to clarify how customers authorize autonomous systems and who is liable if a transaction fails.
How AI Agents Reached Bank Logins After Meta’s Muse Launch
Meta launched Muse, its personal shopping agent, on Sept. 8. Within two weeks it had three checkout rails: Stripe’s Link at launch, Shop Pay, and PayPal. I covered the Shop Pay integration as it landed. The PayPal rail got its own writeup, linked further down.
Part of why this moved fast is that the agent itself changed. Early shopping assistants recommended products and sent you to a checkout page. The current generation fills the cart, picks the shipping method and pays. Once software can pay, every party that guards a payment rail has to decide whether it trusts that software, and the banks have now answered for their own rails.
Gatekeepers started pushing back almost immediately. Amazon blocked Muse on Sept. 21, saying the agent lacked authorization and stored customer credentials. Meta disputed that. A day later a researcher disclosed a flaw that could let a local app hijack Muse’s authentication token, and Meta shipped a hotfix the same day. My writeup of the Shop Pay security hole has the details.
The payments industry then built its own guardrails. On Oct. 14, American Express introduced an intent check for AI agent purchases, which I broke down in this post. Shopify opened its catalog to personal agents the same day, per my coverage of the catalog change.
Mercury took the opposite approach to the banks. According to posts on X by the company’s founder, Mercury launched actions on its MCP server on Oct. 15. Per Mercury’s help center, a connected AI tool can read accounts and transactions, and it can request transfers and payments to recipients already saved in Mercury. Nothing moves until the account holder approves in Mercury, and “selecting Confirm Request in your AI tool doesn’t authorize the payment.” Approval requests expire after about five minutes. Simon Taylor’s Brainfood newsletter drew the same contrast on Oct. 18.
There is a fair counterpoint to the safety framing. The banks have an obvious financial reason to keep agents away from their deposits, and a Simon-Kucher senior partner quoted in the reporting argues banks should respond by making deposits stickier, for example with loyalty programs tied to borrowing rates. Both things can be true: the blocks protect customers from credential-sharing risk, and they protect bank margins.
What the Bank Blocks Mean for High-Ticket Checkout and Chargebacks
My read is that most of this does not touch your storefront today. The restrictions target agents that log into a bank account with the customer’s own credentials. In my earlier coverage, Muse checks out through Shop Pay, PayPal and Link, and Shopify uses single-purchase “human buyer” credentials so Muse never holds a card number. Those tokenized wallet rails do not depend on a buyer’s bank letting an agent sign in.
Where it does bite is the edge. High-ticket buyers pay by bank transfer, ACH, wire and financing more often than a $40 impulse buyer does. If an agent tries to close a $6,000 sale by logging into the buyer’s bank, and the bank blocks it, that order dies at the last step with no error you will ever see. Treat any pay-by-bank option on your store as the exposed rail.
The bigger issue is liability, and nobody has answered it. Bank of America is still writing its protocols, and regulators told the FT the law may need to clarify authorization. Until it does, the merchant is the easiest party to chase. Here is hypothetical math, not a reported figure: 100 agent-placed orders a month at a $4,000 average is $400,000 in volume. If 1% become disputes, that is $40,000 in contested revenue before fees, and the evidence trail for “a software agent bought this” is thinner than for a human who typed a card number.
Visa already widened what counts as evidence. Its Compelling Evidence rules apply to disputes processed on or after Oct. 24, and my post on those evidence rules notes that agentic payment provider login IDs now count as account elements. That helps you win some fights. It does not tell you who pays when the agent was wrong.
Two scenarios to watch. If the Fed or Congress settles agent authorization soon, banks may open controlled doors, as Schwab’s read-only plan suggests, and pay-by-bank agent orders become workable. If not, expect more gatekeepers to block more agents, and expect your checkout mix to matter more. Either way, one quarter of bank statements is not a trend. The next signal is whether Wells Fargo, the third big deposit holder in the FT’s math, takes a public position.
Think about who actually carries the risk on each side. The bank carries it on its own deposits, so it can simply say no. The agent provider, Meta in Muse’s case, carries it on its brand and has already shown it will patch fast when a flaw surfaces. The merchant sits at the end of the chain with the product shipped and the money in question. That ordering is why I treat liability as a store-level problem even though the headlines are about banks.
There is also a timing point worth noting. Visa’s evidence rules take effect Oct. 24, which is four days from today, while the banks are still drafting their own protocols. Rules for the people who accept payments are arriving before rules for the people who hold the money. If you run high-ticket orders, that gap is the period to document everything: the agent’s identity, the session, the delivery confirmation, and the buyer’s stated intent.
The other cost is data. When an assistant completes checkout off-site, you lose browsing history and email capture, the problem I flagged in the Muse coverage. A buyer you cannot email is a buyer you cannot sell a $1,200 accessory to next month.
If wiring all this together sounds like a second job, it is. My team builds and runs stores with the payment, fraud and email layers already in place through the turnkey done-for-you service.
AI agents will shop your store whether you planned for them or not, so do you want 1-on-1 help building a high-ticket store that holds up? Get the coaching details →
Five Steps to Take Before Agent Orders Reach Your Store
None of this needs a rebuild. Here are five moves I would make this week, in order.
- Audit your payment methods in Shopify under Settings, then Payments. List every rail an agent could use: Shop Pay, PayPal, cards, any pay-by-bank option, any financing. If a method depends on the buyer logging into their bank, mark it as the one that can fail silently.
- Set a manual-review threshold for orders that arrive through AI channels. I would hold anything over $2,500 for a human check before releasing it to the supplier. For automated screening on high-ticket orders, ClearSale is the fraud tool I point store owners to.
- Lock down your own banking. Never give an AI tool your bank password. Use an OAuth connection with approval on every payment, which is the model Mercury describes in its security guide, and revoke any connection you stopped using. My walkthrough of opening a Mercury account covers setup.
- Add a second rail so one decision by one bank cannot stall you. I use the single point of failure logic from my backup plan guide, and a multi-currency account like Airwallex gives you a second home for money.
- Rebuild the data you lose to off-site checkout. Put email capture before the cart in Klaviyo, and route post-purchase questions from unknown buyers into Gorgias. Then book a free discovery call and bring six months of disputes if you want a second set of eyes on your evidence file.
Keep your books clean while you do this. Every agent-initiated payment and every transfer you approve should land in a ledger, and Finaloop handles that for Shopify stores without a bookkeeper. If you use Claude to prep that work, give it read access only. Mercury’s own guide warns that read access exposes balances, transaction history and routing numbers, so treat it as sensitive.
Frequently Asked Questions
Do the bank blocks stop Meta’s Muse from buying on my Shopify store?
Not according to the reporting. The blocks cover agents that log into a bank account and move money. In my coverage, Muse checks out through Shop Pay, PayPal and Link, which are tokenized wallet rails.
Who pays if an AI agent buys the wrong item and the buyer disputes it?
Nobody has settled that. Bank of America is still drafting liability protocols, and regulators told the FT the law may need to clarify it. Until then, treat it as any card-not-present dispute and follow my chargeback prevention guide.
Should I connect my business bank account to ChatGPT or Claude?
Only if the setup requires approval inside your bank for every payment, and never with your password. Mercury’s help center says its AI connections cannot move money until you approve in Mercury, and my Mercury pricing breakdown shows what the account costs. This is general information, not financial or legal advice.
Will the banks keep blocking agents?
Unknown. Schwab says it is building read-only access, and Citi allows read-only data sharing through open banking providers. The reported trend is toward controlled access, not a permanent ban.
Does this affect my processor payouts?
Nothing in the reporting touches merchant payouts or settlements. Your payout path is your processor sending money to your bank, which is a different thing from an agent logging into a consumer account.
Should I offer pay-by-bank on a high-ticket store?
Test it with a small share of orders and keep cards and Shop Pay as the default. My high-ticket dropshipping guide covers why order value changes how buyers want to pay.
I do not have a store yet. Where do I start?
Grab the free high-ticket niches list first. Then read my breakdown of the best niches before you pick a product category.
Want my team to scale the store you already have while you sort out agent-ready payments? See the scaling service →
I will keep watching which banks move next and what Shopify does about agent payments. Subscribe to the YouTube channel for daily breakdowns. More breaking news coming through the day.
Related Articles
If this was useful, these go deeper:
- No Shopify Store in 122 Passed All 9 AI-Agent Checks
- How to Build a US E-Commerce Business From Abroad Without Creating a Banking Mess
- Treasury Yield Hits 24-Year High. Check Your Credit Lines
- Protecting Your High-Ticket Store: Security Infrastructure for Scaling
- Claude for Ecommerce: How to Build a Leaner High-Ticket Business With AI

Trevor Fenner is an ecommerce entrepreneur and the founder of Ecommerce Paradise, a platform focused on helping entrepreneurs build and scale profitable high-ticket ecommerce and dropshipping businesses. With over a decade of hands-on experience, Trevor specializes in high-ticket dropshipping strategy, niche and product selection, supplier recruiting and onboarding, Google & Bing Shopping ads, ecommerce SEO, and systems-driven automation and scaling. Through Ecommerce Paradise, he provides free education via in-depth guides like How to Start High-Ticket Dropshipping, advanced training through the High-Ticket Dropshipping Masterclass, and fully done-for-you turnkey ecommerce services for entrepreneurs who want a faster, more hands-off path to growth. Trevor is known for emphasizing sustainable, real-world ecommerce models over hype-driven tactics, helping store owners build scalable, sellable, and location-independent brands.

