Citi bank branch facade with blue signage and logo displayed above entrance

Citi shifts AI fraud risk onto customers as banks redraw the rules for finance agents

Citi now makes customers responsible for losses from AI finance tools, joining JPMorgan, Bank of America and PNC.

In short

Citi has changed its online terms to make customers liable for losses tied to third-party AI finance tools, including when those tools are hacked. The move follows similar policy changes from other major U.S. banks and highlights a growing fight over risk in AI-driven banking.

  • Citi updated its online user agreement this week to shift AI-tool loss responsibility to customers.
  • The policy covers third-party AI finance tools, including cases where credentials leak or the tool is hacked.
  • JPMorgan Chase and Bank of America adopted similar policies last month, and PNC changed its terms today.
  • Banks see AI agents as both a security risk and a competitive threat in personal finance.

Citi has updated its online user agreement to make customers responsible for losses tied to third-party AI finance tools, including cases where those tools are hacked. The move, reported by TechCrunch, adds fresh pressure to a fast-growing fight over who pays when an AI agent moves from advice to action inside a bank account.

What Citi changed

The new language shifts responsibility to customers who give third-party AI agents access to their accounts. That covers personal finance AI tools that use bank credentials, and it extends to intentional or accidental transactions carried out by AI assistants.

According to TechCrunch, Citi’s updated terms go a step further by saying customers bear losses even if an AI tool leaks credentials or is hacked. The bank’s stance is straightforward: if the customer authorizes the tool, the customer carries the risk.

“we want customers to understand the risks associated with sharing banking credentials or authorizing those tools to access their accounts”

Why it matters

This is not a narrow terms-of-service tweak. It is a sign that large banks are trying to redraw the legal and practical boundaries around AI agents in consumer finance before the technology becomes mainstream.

AI tools are increasingly being pitched as personal finance helpers that can analyze spending, suggest moves, and in some cases execute transactions. That creates an obvious tension. The same automation that makes the tools useful also makes them harder to supervise, easier to misuse, and more complicated to assign blame when something goes wrong.

For customers, the immediate issue is liability. If a third-party tool is compromised, makes an unauthorized purchase, or triggers an error after a user has granted access, Citi is making clear that the bank does not want to absorb the loss by default.

For the banking industry, the issue is broader. As AI agents become more capable, banks face two risks at once: security exposure and competitive pressure. If banks do not set their own rules, AI firms may end up owning the customer relationship and the transaction flow, along with the fees that come with it.

How other banks are responding

Citi is not moving alone. TechCrunch reported that JPMorgan Chase and Bank of America adopted similar policies last month, and PNC changed its terms today in the same direction. Citizens Financial Group also emailed customers this week to warn that AI assistant actions are the customer’s responsibility.

The speed of those changes suggests a coordinated industry reaction rather than a one-off policy update. The largest banks are setting a baseline: if a customer connects a third-party AI agent to a bank account, the customer may be taking on a significant share of the risk.

Bank Reported change Timing Size note from TechCrunch
Citi Updated online user agreement to make customers responsible for losses tied to third-party AI tools, including hacks This week Third largest U.S. commercial bank
JPMorgan Chase Adopted a similar AI-agent liability policy Last month One of the two largest U.S. banks
Bank of America Adopted a similar AI-agent liability policy Last month One of the two largest U.S. banks
PNC Financial Services Changed its terms in the same direction Today Eight largest U.S. bank
Citizens Financial Group Emailed customers warning that AI assistant actions are their responsibility This week Not specified in the fact sheet

How does this affect third-party AI finance tools?

It makes them harder to sell as low-friction consumer products. If a tool can see account data, suggest transfers, or place trades, the user is no longer just getting software recommendations. The user is effectively authorizing a digital intermediary to act inside a sensitive financial relationship.

That matters because personal finance AI is still a young product category. Startups are pursuing users and transaction-related fees by promising convenience, automation, and faster decision-making. But the more control they gain, the more likely banks are to treat them as a security and liability issue rather than a simple app integration.

TechCrunch noted that Anthropic’s personal finance product is read-only and cannot move, manage, or invest money. That distinction is important. Read-only tools can still help people organize information, but they stop short of the kind of account control that triggers the liability questions now confronting the banks.

Read-only versus transaction-capable tools

A read-only finance assistant can scan balances, summarize spending, and surface trends without changing the account itself. A transaction-capable tool can go further, which is where the legal and operational stakes rise sharply.

  • Read-only tools can analyze and explain.
  • Transaction-capable tools can move money or make purchases.
  • Bank liability terms now appear designed to limit losses from the second category.

Background: how banks and AI agents got here

AI agents can be given access to bank and brokerage accounts to analyze spending, suggest actions, and sometimes execute transactions. That promise is what has drawn both banks and startups into the same market, even as they appear to want very different outcomes from it.

Banks already operate under strong authentication requirements and established rules around unauthorized activity. In ordinary fraud disputes, those protections can shape who absorbs the loss. AI tools complicate that picture because the customer may have technically authorized access, even if the resulting action was never intended or was taken without full understanding of the consequences.

That is the core of the dispute now emerging. If a customer uses a third-party AI agent as a financial assistant, is the bank still responsible when the agent makes a bad move? Or does the customer own the outcome because they turned the tool loose on the account?

The banks’ answer, at least for now, is increasingly leaning toward the second option.

Who is pushing AI deeper into personal finance?

AI firms are pushing agents into personal finance to win users and fees. That is the commercial engine behind the product push, and it explains why the issue is moving so quickly from technical experimentation to bank policy.

For startups, finance is an attractive market because users return frequently and transactions can generate recurring revenue. For banks, that same attraction raises alarms. If a third-party agent becomes the place where a customer plans, acts, and pays, the bank may be reduced to a utility in the background.

That competitive threat helps explain the hard line in Citi’s terms. A bank that loses control over the customer interface may also lose control over the risk narrative. By updating its agreement now, Citi is trying to define the legal perimeter before AI finance assistants become a normal part of customer behavior.

What Citi’s terms suggest about bank strategy

Citi’s language suggests a two-track strategy: permit AI use, but only on the bank’s terms. The company says customers can use Citi’s own AI agents without added liability, while third-party tools trigger a different risk framework.

That distinction matters. It shows Citi is not rejecting AI outright. Instead, it is trying to keep customers inside tools the bank can supervise more closely, while making outside tools a customer-driven decision with customer-owned consequences.

The policy also reflects a common banking instinct: preserve the security perimeter. Once a customer authorizes a third-party service to access account credentials or activity, the bank’s ability to control the transaction chain weakens. The new terms are designed to close that gap by contract, if not by technology.

By the numbers

The reported changes land at a moment when the biggest U.S. banks are setting the pace for the industry.

Data point Value Meaning
Citi rank Third largest Shows the policy comes from a major national bank, not a niche lender
JPMorgan Chase / Bank of America rank Two largest Indicates the biggest players moved first
PNC rank Eight largest Suggests the policy spread beyond the very top tier
Policy timing across banks Last month to today Shows a rapid industry response over a short window

What regulators may watch

Regulators are likely to pay attention to whether banks can use terms of service to shift responsibility in situations that consumers may still view as unauthorized or fraudulent. The fact sheet does not identify any regulatory action, but the policy move raises obvious questions about how existing consumer protection rules apply when an AI agent is acting under user permission.

That leaves several open questions. If a customer grants access to an AI tool that is later hacked, is the customer’s authorization enough to shift the loss? If the tool makes an accidental transfer, is that treated differently from a deliberate one? And if the bank approved the connection structure but not the specific action, where does the blame sit?

Those disputes could become more common if AI finance products move beyond monitoring and into execution. Banks may prefer to settle that question now, before a broader wave of customer complaints forces the issue into public view.

What happens next

The immediate next step is likely more policy updates. TechCrunch reported that multiple major banks have already adjusted their terms within a short period, and that pattern could continue if more institutions decide they need the same liability shield.

What to watch:

  1. Whether more major banks adopt similar AI-agent liability language.
  2. Whether customers challenge losses tied to hacked or malfunctioning AI tools.
  3. Whether regulators question contract language that shifts liability to users.
  4. Whether AI finance products stay read-only or move toward transaction execution.

For now, the message from Citi and its peers is clear: AI may be entering consumer finance, but the banks do not want to be the first line of payment when it goes wrong.

Context: the new trust problem in consumer finance

The rise of AI agents in banking is creating a trust problem that is part technology, part contract, and part consumer expectations. Customers want convenience. Banks want control. AI firms want distribution and fees. Those three goals overlap, but they do not align neatly.

That is why this story matters beyond one revised user agreement. It shows how quickly financial institutions are moving to define the rules around an emerging product category before it becomes normal enough to feel invisible.

The policy fight is just beginning, but the direction is already visible: as AI agents get closer to the money, the banks want the liability line drawn as close to the customer as possible.

Frequently asked questions

Why are banks changing their terms for AI finance tools?

Banks are changing their terms because AI finance tools can access accounts, suggest transactions, and sometimes execute them, creating new security and liability risks. By updating user agreements, banks are trying to make clear that customers bear the loss if they authorize third-party tools.

Does Citi make customers responsible if an AI tool is hacked?

Yes. Citi’s updated terms say customers are responsible for losses tied to third-party AI tools even if the tool leaks credentials or is hacked, according to TechCrunch.

Which other banks have similar AI liability policies?

JPMorgan Chase and Bank of America adopted similar policies last month, PNC changed its terms today, and Citizens Financial Group warned customers this week that AI assistant actions are their responsibility.

Can AI finance tools move money from bank accounts?

Some can, but not all. TechCrunch said Anthropic’s personal finance product is read-only and cannot move, manage, or invest money, which avoids the transaction-risk issues that banks are now trying to control.

What should customers watch next?

Customers should watch for more banks adopting similar liability language, possible disputes over fraudulent transactions, and any regulatory response to banks shifting risk onto users of third-party AI tools.

Share this 🚀