
Wells Fargo's new online agreement takes effect November 24, 2026, and Citi and PNC changed terms this week. Banks are writing customer liability for third-party AI agents into contracts before any law says who should carry the risk.
The first rule of connecting an AI agent to your bank is that the bank has already decided who pays when it goes wrong, and the answer is you. In a month, several major US banks have rewritten their customer agreements to say so. Wells Fargo's takes effect November 24, 2026. Citi, PNC and Citizens changed terms this week, and JPMorgan and Bank of America announced similar policies in October, according to TechCrunch. There is no federal statute in the reporting we reviewed that settles the question, so for now the contract is the law that matters.
What Wells Fargo's notice actually says
The primary document is Wells Fargo's own update page. It tells customers: "If you continue using these services after November 24, 2026, the updated Agreement applies." Its summary of the new Section 15 says that if you use a third-party AI tool or agent to access your accounts, the tool is not operated by Wells Fargo, you authorize the bank to share the information the service needs, requests made through the tool may be treated as coming from you, and you are responsible for managing its access. The bank reserves the right to restrict or block certain third-party AI tools. Separate sections on authorized users and delegates say that if you let someone else access your accounts you are responsible for their actions, including actions taken through Wells Fargo's own AI-powered tools.
We read the summary page, not the full agreement text, which Wells Fargo hosts on a separate page, so the exact wording of the section is not quoted here.
American Banker reported on November 2 that, starting November 24, customers who tell an agent to stop accessing their accounts "would still be liable for any agent activity unless they produce evidence they did so." It called Wells Fargo the apparent first bank to require customers to prove they have turned off an agent's access. The article does not say what proof the bank will accept, a gap Wolters Kluwer's Elaine Duffus raised when she asked, "what happens then?"
“If you continue using these services after November 24, 2026, the updated Agreement applies.”
The other banks
TechCrunch's November 18 report names the rest. Citi, the third-largest US commercial bank, updated its Online User Agreement this week. A Citi spokesperson said the bank "recently updated our Online User Agreement," and that customers "are responsible for activities conducted through those agents." The update also covers hacks: if an AI tool leaks credentials or is hacked, the customer is responsible for what the hackers do with the account. Banks say they will not reverse transactions made by automated assistants, intentional or not. PNC changed its terms on November 18, and Citizens emailed customers warning that they are responsible for anything their AI assistant does. Agents that Citi provides itself are excluded from the liability language.
The October moves were framed differently. PYMNTS reported on October 19, 2026 that between October 13 and 16 banks restricted outside agents. At JPMorgan, outside agents can see balances through an authorized connector such as Plaid but cannot send money or pay bills. Charles Schwab blocks unauthorized outside agents, including Meta's Muse, which CEO Rick Wurster said it sees "in the tens and 20s" each week. Bank of America limits agent access to consumer deposits while it writes liability rules, and a human should approve money leaving an account. JPMorgan CEO Jamie Dimon called outside agents "rogue agents" and described the customer's position as "the liability transfer."
Why this is a safety story, not just a legal one
The banks' position rests on a technical fact: an agent that holds your login is indistinguishable from you. That is the same access problem behind the privacy failures Reuters reported in Meta's Muse, which we cover in our piece on what Reuters found in Meta's Muse testing. It is also why the contractual line sits where it does. If an agent is tricked, hallucinates a payment or is hijacked, the bank cannot tell it from you and does not want to eat the loss.
Anthropic's personal finance tool, launched in early November 2026, takes the opposite design. TechCrunch describes it as "read-only," unable to "move, manage, or invest your money." Read-only access is the safest setting to ask any agent for, because it removes the transaction risk the contracts are shifting.
What is not known
The sources we reviewed do not show widespread unauthorized transfers or customer losses from agents. They also disagree on details. Citi's October statement said it does not let third-party agents log in or transact, while its November update describes customer responsibility for agents a customer authorizes. The reporting does not reconcile the two, and we did not open Citi's agreement. TechCrunch's article does not mention the Consumer Financial Protection Bureau or Regulation E, the federal rule on electronic transfers, so we cannot say how regulators would treat these clauses.
What an agent holder can check today
Three practical checks follow from the documents. First, find the section on AI tools in your bank's online agreement and read whether requests through a tool are treated as yours. Second, find out whether your bank offers an official read-only connection, which is how JPMorgan lets outside agents see balances. Third, if you stop using an agent, write down the date and keep proof, because Wells Fargo's reported rule puts the burden of showing revocation on the customer.
Our take
These clauses are a negotiating position as much as a risk policy. Banks are asking customers to accept liability for software the banks do not control, from vendors the customers cannot audit, with no standard for what counts as proof of revocation. We would not connect a write-capable agent to a bank login before November 24 or after. If you use an agent at all, give it read-only access through an official connector, keep a record of the day you revoked it, and ask your bank in writing which agents it permits. Watch for a regulator or court to say whether a contract can shift liability for unauthorized transfers this way.
Frequently asked questions
Who is liable if my AI agent makes a bad transaction at my bank?
Under the terms several banks have published or updated, you are. Wells Fargo's agreement from November 24, 2026 says requests through an AI tool may be treated as coming from you. Citi says customers are responsible for activities conducted through agents, including after a hack.
When does Wells Fargo's new AI agent rule take effect?
November 24, 2026. Wells Fargo's notice says continued use of its online services after that date means the updated Online Access Agreement applies, including a new section on third-party AI tools and agents.
Which banks have changed their terms for AI agents?
Per TechCrunch, Citi, PNC and Citizens changed terms in the week of November 18, 2026, and JPMorgan and Bank of America announced similar policies in October. Wells Fargo's agreement takes effect November 24, and PYMNTS reports Charles Schwab blocks unauthorized outside agents.
Can I prove to my bank that I revoked an AI agent?
American Banker reports Wells Fargo will require customers to produce evidence that they turned off an agent's access to avoid liability for its activity. The article does not say what form that evidence must take or how the bank will assess it.
Is it safe to give an AI agent access to my bank account?
Read-only access is the lowest-risk setup. Banks such as JPMorgan allow outside agents to see balances through authorized connectors like Plaid but not to send money or pay bills. Giving an agent your login credentials leaves you carrying the liability under current terms.
Is there a law that says who pays when an AI agent is hacked?
None of the sources we reviewed identifies one. TechCrunch's reporting does not mention federal consumer-finance rules, so the contract terms are what govern for now. Whether such clauses can shift liability for unauthorized transfers is untested.
Sources
What each one is, and whose it is.
- 1
We are updating our Online Access Agreement, Wells Fargo (November 20, 2026)
Documentation - 2
More banks say customers using AI agents are on their own, even if they get hacked, TechCrunch (November 18, 2026)
Press reportIndependent of the vendor - 3
Wells Fargo wants customers to prove they've revoked AI agents, American Banker (November 2, 2026)
Press reportIndependent of the vendor - 4
Big banks won't let outside AI agents move money, PYMNTS (October 19, 2026)
Press reportIndependent of the vendor