Construction AI BriefSubscribe →
Issue
№456
Pillar
Trend
Audience
Trade sub
Dated
2026.11.20

Big banks now say losses from an AI agent using your login are yours. Subs letting AI pay supplier bills should check the bank agreement first

Citi, PNC and Citizens joined JPMorgan, Bank of America and Wells Fargo this month in putting AI agent activity, including hacks, on the customer. For a trade sub whose office is testing agents on supplier payments and bank reconciliation, the rule is simple: no agent gets the bank login until someone has read the agreement.

ByConstruction AI BriefAbout this publication

Six large U.S. banks now say that if you hand an AI agent your online banking login, whatever it does is on you, and at Citi that includes what a hacker does if the AI tool leaks your credentials. For a mechanical or electrical sub whose office manager is testing an agent to pay supplier invoices or reconcile the operating account, that settles one question. Until you've read your own bank agreement, no AI agent gets the bank login.

What did the banks change?

Per TechCrunch, Citi revised its Online User Agreement this week. Customers who give access to third-party AI tools are "responsible for activities conducted through those agents." Citizens emailed customers saying much the same, PNC changed its terms on November 18, and JPMorgan and Bank of America announced similar policies last month. The banks also say they won't be responsible for reversing transactions an automated assistant makes, whether it meant to or not.

Here's where the large banks stand, based on the reporting so far:

BankWhat's been reported
CitiCustomer is responsible for agent activity, including hacks through a leaked AI tool. Agents Citi provides itself are the exception.
Wells FargoUpdated Online Access Agreement takes effect Nov. 24. Agent instructions are "deemed to have been initiated and authorized directly by you, without further verification."
Bank of AmericaActivity from AI access is the customer's responsibility. A human must approve money leaving the account.
JPMorgan ChaseExecutives say customers who share credentials with an outside agent carry the risk.
PNC, CitizensTerms or customer notices updated this week. Details are thin.

Wells Fargo's Section 15 defines an AI agent as "any software that takes autonomous, semi-autonomous, or non-autonomous action on behalf of any person or entity." That's broad enough to cover a browser agent, a desktop assistant or a homemade script.

Why does this matter to a trade sub now?

In the same week, the tools that could log into a bank portal got easier to switch on. Google extended Gemini's computer use to the Mac for paid U.S. users, and Microsoft put its always-on Copilot Autopilot agent into early access. A controller who wants to stop keying 40 supplier payments a week can now point an agent at a browser and walk away. The bank has just said, in writing, who owns the result.

There's a gap to be honest about. None of the coverage says whether these terms reach business or treasury accounts. They describe consumer online banking. Don't read that as a loophole. The federal consumer protections mentioned in the coverage come from Regulation E, which defines an account as one "established primarily for personal, family, or household purposes." Your operating account isn't one of those, so it lives on whatever your commercial agreement says. Plenty of small subs also run some company bills through an owner's personal account or card, and that money falls under exactly these new consumer terms.

What should the office do this week?

  1. Ask who has given an AI tool a bank login. Check the office manager, the controller, the owner and any bookkeeper or outside accountant. Include browser extensions and "connect your bank" features in budgeting or AP apps.
  2. Pull the business banking or treasury agreement and search it for "agent," "third party," "credentials" and "automated." If it says nothing, email your banker and get the answer in writing.
  3. Keep agents on the accounting side of the line. An agent can read supplier invoices, match them to POs and receiving tickets, flag price changes against the quote, and stage a payment batch in your accounting system. A person with their own login releases the batch in the bank portal.
  4. Use the bank's own controls. Dual approval on ACH and wires, Positive Pay, and per-user limits are already in most business banking packages. They also catch the ordinary fraud a sub sees, like a changed remit-to address on a supplier invoice.
  5. Rotate the password on any bank login that has ever been pasted into an AI tool.

What's still unclear?

How the banks will apply these terms to small business accounts, and whether bank-approved agent connections will arrive soon enough to make this moot. Citi already exempts agents it provides itself. A bank-sanctioned, read-only connection for reconciliation would be a very different risk from an agent holding the owner's password.

We covered the same "agent with its own access" problem from the PM side in our piece on Microsoft's Autopilot agent. The takeaway for a sub's back office is short. AI can prep the payment run. A person releases the money.

Construction AI Brief covers what the week's AI news means for GCs, subs and estimators, three times a week. Subscribe at constructionaibrief.com.

FAQCommon questions
Am I liable if an AI agent with my bank login makes a bad payment?
At several of the largest U.S. banks, yes. Citi, Wells Fargo and Bank of America have updated their online terms so that actions an AI agent takes with your credentials count as authorized by you, and Citi's terms say that includes what hackers do if the AI tool leaks your credentials.
Which banks have changed their terms on AI agents?
As of November 2026, reporting covers Citi, PNC and Citizens (updated this week), JPMorgan and Bank of America (similar policies last month) and Wells Fargo, whose updated Online Access Agreement takes effect November 24, 2026 and says use of any AI agent is 'AT YOUR OWN RISK.'
Do these bank AI agent rules apply to business accounts?
The reporting so far describes consumer online banking terms and does not say whether business or treasury accounts are covered. A contractor should read its own business banking or treasury management agreement, or ask its banker, before giving any AI tool access.
Does Regulation E protect a construction company from AI agent fraud?
Regulation E's definition of 'account' covers accounts established primarily for personal, family or household purposes, so it is a consumer rule. A business operating account is governed by its own bank agreement, not Regulation E's consumer liability limits.
Can a trade contractor still use AI for accounts payable?
Yes, but keep the agent out of the bank. Let it read invoices, match them to POs and draft the payment batch in your accounting system, and have a person with their own credentials release payments in the bank portal.
End of sheet — issue №456
Published · 2026.11.20
Project
Construction AI Brief
Dated
2026.11.20
Sheet
1 / 1
Rev
A
Published independently · constructionaibrief.com · © 2026Facebook·Privacy·About