Policy & Regulation

Four banks make customers liable for third-party AI agent losses

Citibank, JPMorgan, Bank of America and PNC have shifted responsibility for losses caused by third-party AI agents to customers. Citi's terms reach hacker activity enabled by shared credentials.

Bank Customers On the Hook if Third-Party AI Leads to Losses
Bank Customers On the Hook if Third-Party AI Leads to LossesStartupStockPhotos / pixabay
By Rebecca Stone6 min read

Why it matters

  • A TechCrunch report dated Nov. 18, 2026 identified four banks that changed their terms: Citibank, JPMorgan, Bank of America and PNC.
  • Citi says customers are responsible for activities conducted through third-party AI agents they authorize.
  • Citi's reported terms cover anything hackers do with credentials or login information acquired through third-party-agent use.
  • One airline, United Airlines, recently banned third-party agents from booking flights.

By Rebecca Stone

Four banks—Citibank, JPMorgan, Bank of America and PNC—have made customers responsible for losses tied to third-party AI agents, according to a Nov. 18, 2026 TechCrunch report cited by WebPro News. The shift matters because some customers have given those agents their banking credentials so the tools can log in and perform actions inside their accounts.

WebPro News said banks were “horrified” to discover customers were using third-party AI agents on their behalf, including handing over credentials so the tools could log in. The practice gives banks a security problem after a customer has authorized an outside service to operate inside a financial account.

What changed in the banks’ terms?

The TechCrunch report identifies four banks that have amended their terms: Citibank, JPMorgan, Bank of America and PNC. It says customers now bear responsibility when third-party platforms impose costs on the bank or leave them exposed to hacking.

WebPro News characterized that policy direction as protecting the banks before protecting customers. The source did not reproduce the banks’ full agreements or quote each institution’s clause. It also did not specify when every change took effect.

Citi’s updated Online User Agreement provides the clearest description in the material. It places responsibility for account activity conducted through a third-party AI agent on the customer who grants access.

What does Citi’s language cover?

A Citi spokesperson told TechCrunch:

“As third-party AI tools become more widely available, we want customers to understand the risks associated with sharing banking credentials or authorizing those tools to access our accounts,”

“We recently updated our Online User Agreement to provide greater clarity about those risks and to reinforce that customers who choose to grant account access to third-party AI agents are responsible for activities conducted through those agents.”

The two quoted sentences address related duties. First, Citi says customers are responsible for activities conducted through agents they authorize. Second, WebPro News said Citi’s terms make customers responsible for “anything” hackers may do with credentials or login information acquired because the customer used a third-party agent.

The cited language reaches beyond activity conducted inside the account by the agent. It also addresses a later hacker who obtains the same credentials or login information. In the sequence described by the source, the customer first shares credentials, an agent uses them, and hackers later acquire the same information. Under Citi’s reported terms, the customer remains responsible for the hackers’ activity.

The source does not describe each bank’s agreement in identical language. The broader reported policy is contractual: four banks have said customers choosing third-party agents bear the resulting financial exposure.

For consumers, the operative issue extends beyond whether an agent completed an allowed transaction. It includes whether granting access transfers responsibility for later credential misuse. Citi’s quoted answer is yes.

Why are banks drawing a hard line?

WebPro News said banks are grappling with how to protect themselves and their customers after third-party services gain access with valid credentials. Those goals now pull apart whenever an outside agent acts with that access. The new terms resolve the tension by assigning financial responsibility to the customer who granted it, according to the source’s description.

The financial stakes are direct: a customer can authorize an agent and still absorb a loss caused by the resulting credential exposure. For banks, the changed terms place that exposure on the customer rather than leaving it with the institution.

The report describes two potential paths for loss after access is granted. One runs through activity conducted by the agent. The other runs through hacker use of credentials obtained after the customer connected the agent. Citi’s language addresses both, according to WebPro News.

The reported behavior matters because the agents could act. They did not merely draft instructions for a customer to approve; the report says they logged in and performed various actions. A term built around authorization must therefore cover more than a recommendation.

That makes credential sharing central rather than incidental. The cited Citi language attaches responsibility to granting access, including access that later exposes credentials to hackers. It does not narrow the described hacker-loss provision to a specific command given to an agent.

Is this only a banking problem?

No. WebPro News cited United Airlines, which recently banned third-party agents for flight bookings. The airline rule addresses a different transaction but the same permission problem: an outside agent can act for a customer once the customer supplies access or authorization.

The comparison shows that banks are not alone in reconsidering how third-party AI tools connect to consumer accounts. It does not establish that other airlines or financial institutions have adopted the same restrictions. The source named only United Airlines.

United has prohibited a specific use of agents. Banks have taken a different route: they address third-party access through terms that allocate resulting losses to customers. One limits a use; the other assigns responsibility.

What remains unresolved?

The material does not include the complete agreements for Citibank, JPMorgan, Bank of America or PNC. It also does not say whether the banks use identical language or apply the same exceptions.

The report does not state whether the clauses include liability caps, carve-outs for unauthorized activity or different treatment for different forms of access. It therefore establishes the policy direction but not the legal result of a particular claim.

The cited material gives no effective dates for the bank agreement changes. A customer could encounter a new online agreement without knowing from the report when responsibility began for an earlier incident. The source does not say whether any bank would apply updated terms retroactively.

The unresolved issue is access control. By handing over credentials, a customer can turn a third-party tool into a route into the account. A hacker who acquires those credentials can use the same route, and Citi’s reported terms leave the customer facing the loss.

Who bears the loss now?

Under the four banks’ reported policy, the customer bears the loss when third-party access causes costs or a hack. Citi provides the broadest description in the source: its terms make customers responsible for anything hackers may do with credentials or login information acquired as a result of using third-party agents.

This is a contractual change, not evidence that every customer accepted the risk in practice or that banks stopped trying to stop fraud. The source does not report changes to monitoring, reimbursement rules or security controls.

WebPro News’s conclusion was blunt: “Regardless of how tempting it may be to use third-party AI agents for banking, customers do so at their own peril.” The bank terms turn that warning into a stated allocation of responsibility for agent-enabled account activity.

The practical result is already clear: a customer who grants an AI agent account access can remain liable for what hackers do with the resulting credentials, and any narrower protection would have to come from the full agreements the cited report did not publish.

Source: webpronews.com

Share this article:

More from Rebecca Stone

Rebecca Stone

Show full bio

Correspondent covering companies, deals and strategy in artificial intelligence and machine learning at AI In Context.

754 articles

Related articles

  1. Bank of America warns customers not to share passwords with AI agents
  2. Citi shifts legal risk of AI agents onto its own customers
  3. JPMorganChase taps OpenAI for Codex and cybersecurity
  4. Uptiq launches GovernAI to govern AI agents in banking at no extra cost
  5. Wall Street Is Pricing In an AI Banking Panic. Bank CEOs Aren't Buying It

« Previous articleNext article »