# Five US Banks Now Make Customers Liable for Outside AI Agent Losses

- URL: https://usa-24.com/ai-agent-liability-bank-terms/
- Publisher: USA 24
- Section: Artificial Intelligence (AI)
- Published: 2026-11-21T08:37:54.098Z
- Updated: 2026-11-21T08:37:54.101Z
- Topics: AI Agents, Online Banking, Consumer Protection, Regulation E, Fraud Liability

> Citi, JPMorgan Chase, Bank of America, PNC and Citizens now hold customers answerable for losses from outside AI agents, even after a hack. Wells Fargo's rule starts November 24.

## Key takeaways

- Citi, JPMorgan Chase, Bank of America, PNC and Citizens have adopted near-identical language that makes customers answer for losses from outside AI agents, according to AI Governance.
- Citi's Section H treats activity carried out with customer-supplied credentials as the customer's own, even if those credentials later leak.
- Wells Fargo's Online Access Agreement takes effect November 24 and treats every delegate transaction, AI agents included, as authorized.
- Banks say they will not reverse transactions an AI assistant initiated, whether those transactions were mistakes or not.
- Whether these clauses hold up under Regulation E has not been tested, and no CFPB or OCC response had been issued as of the coverage.

Five large US banks now put AI agent liability on their own customers, and the rule reaches losses even when the outside agent has been hacked. The change began with Citi and has widened over the past three days. Reporting on November 18 and 19 described PNC and Citizens following Citi, and Wells Fargo's version takes effect on November 24. Anyone who connected an AI assistant to a bank account is the most exposed group, and in several cases the new terms are already in force.

USA 24 covered Citi's change in an [earlier story](https://usa-24.com/citi-ai-agent-terms-liability/), which remains the best starting point for the Citi details. This article relies mainly on secondary reporting. The original TechCrunch report, dated November 18, is the basis for much of the coverage, and the bank agreements themselves were not read directly for this story.

## Which banks have changed their AI agent terms

Across the reporting, five banks are named as adopting near-identical AI-agent liability language: Citi, JPMorgan Chase, Bank of America, PNC and Citizens, according to [AI Governance](https://aigovernance.com/news/five-major-us-banks-disclaim-ai-agent-liability-including-after-hacking). The timing differs from bank to bank.

- Citi rewrote its online user terms so that customers who give an outside AI agent access to their accounts answer for what that agent does, including after a breach. [DigitalToday](https://www.digitaltoday.co.kr/en/view/127747/us-major-banks-warn-on-external-ai-agents-customers-liable-if-problems-arise) relayed the change from TechCrunch.
- JPMorgan Chase and Bank of America adopted comparable language in October 2026.
- PNC reworked its terms on November 18, according to [Startup Fortune](https://startupfortune.com/citi-pnc-and-citizens-tell-customers-they-own-any-loss-an-ai-agent-causes/).
- Citizens sent customers an email in the same week with a similar warning.
- Wells Fargo's Online Access Agreement takes effect November 24, and it works differently from the others, as explained below.

Some dates are not settled. [BigGo Finance](https://finance.biggo.com/news/94a3d1e1-8359-4d23-89eb-699326b4a8a2) refers to a November 24 date for a PNC change, which conflicts with the November 18 date that Startup Fortune reported. The Wells Fargo date is the one most sources agree on. Some outlets also list Charles Schwab and Wells Fargo in the October wave, but USA 24 could not confirm either one in a second independent source.

## What Citi's Section H says about hacked credentials

Citi's added section, labeled Section H, treats activity carried out with customer-supplied credentials as the customer's own, even if those credentials later leak. [Pebblous](https://blog.pebblous.ai/blog/citi-ai-agent-liability-terms/en/) and [TFTC](https://www.tftc.io/citi-ai-agent-liability-customers-section-h) both describe that structure. The practical effect is that a thief who uses a leaked login gets treated the same way as the customer's own agent.

The rule has a carve-out. Citi exempts its own first-party AI agents from the added liability, so the rule covers only outside tools, according to AI Governance. Citi may also restrict, block or cut off a third-party agent's access without notice, and the customer is responsible for monitoring what that agent does, as Pebblous reports.

A Citi spokesperson, who is not named in the source reporting, said that customers who grant an external AI agent access are "responsible for activities carried out through that agent." That statement, relayed by DigitalToday from TechCrunch, is the most direct description of the bank's position in the coverage.

The structure matters because of how it reads in practice. Under the common reading, an AI agent holding your login is treated like you acting yourself. [TechRadar](https://www.techradar.com/ai-platforms-assistants/major-us-banks-crack-down-on-ai-agent-use-by-customers-telling-them-theyre-on-their-own-if-it-gets-hacked-or-makes-a-mistake) describes the bank treating any action a third-party agent takes as authorized by the customer. A mistake by the agent and a theft using stolen credentials land in the same bucket.

## Wells Fargo's November 24 delegate rule

Wells Fargo's approach uses different language. Its Online Access Agreement, effective November 24, treats every transaction by a customer's delegate as authorized, and AI agents fall under that heading. Startup Fortune quotes the agreement as covering AI-agent transactions "including those you did not want or intend." That phrase reaches beyond hacks into ordinary mistakes, which matters for anyone who lets an assistant act on their behalf.

The reporting does not say how Wells Fargo defines a delegate, or whether a customer can carve an agent out of that definition. Customers who use Wells Fargo should read the agreement before the effective date and check whether any existing agent setup falls under it. Because the agreement is framed around delegates rather than outside tools, its reach may be wider than the Citi language.

## Are banks liable if an AI agent gets hacked?

Based on the terms as reported, the customer generally carries the loss, not the bank. The banks' position is that they are not obligated to reverse agent-initiated transactions. DigitalToday reports that the banks say they will not reverse transactions an AI assistant initiated, whether those transactions were mistakes or not.

That position collides with the consumer-protection framework for electronic transfers. Regulation E, the federal electronic-transfer rule, is the central reference point. Startup Fortune flags the open question of whether any regulator will weigh in on how Regulation E treats agent-initiated activity. A separate report says that no response from the Consumer Financial Protection Bureau or the Office of the Comptroller of the Currency had been issued as of that coverage.

Commentators cited by Startup Fortune and TechRadar describe the law as unsettled. As far as the reporting shows, the liability clauses have not been tested in a dispute. The honest answer to the question of whether banks are liable is that the contract language points toward the customer, while the enforceability of that language remains an open question.

## Who carries the risk and what other gatekeepers did

Consumers who linked a third-party AI assistant, budgeting app or agent to a checking, credit or brokerage account face most of the risk, including losses from hacks. The people most exposed are likely those who set up an agent once and then stopped watching it. The terms put monitoring on the customer, so an unattended agent is the weakest point in the arrangement.

Business owners who gave outside AI tools access to business banking should check whether the same terms apply, according to AI Governance. Small companies that handed agents access to payables or card accounts may have the least room to absorb a loss.

The same week, other gatekeepers tightened their rules as well. According to [AI to ROI](https://ai2roi.substack.com/p/ai-to-roi-news-and-analysis-november-78a), United blocked AI agents, joining Amazon and eBay, and JPMorgan blocked agents that handed over raw card numbers. That pattern suggests that companies holding valuable accounts or transactions now treat outside agents as a risk to contain.

The risk is not hypothetical. USA 24 has reported on [GAO covert tests](https://usa-24.com/gao-ai-chatbot-attack-plans/) that got ChatGPT, Claude, Gemini and Grok to plan attacks. It has also covered [California's lawsuit against OpenAI](https://usa-24.com/california-sues-openai-ai-agents/) over 80 lost-control incidents in AI testing.

## Steps to protect your accounts before the terms bite

- Search your bank's online agreement for "AI agent," "third-party access" and delegate language. At Citi, look for Section H.
- Limit outside agents to read-only access wherever the service allows it.
- Where your bank offers its own assistant, consider using it. Citi exempts its first-party agents from the added liability, according to AI Governance.
- Turn on transaction alerts for every account an agent can reach, so unexpected activity shows up quickly.
- Require manual approval for transfers. Bank of America has advised customers to authorize transfers before an AI agent moves funds, according to BigGo Finance.
- Revoke agent credentials you no longer use. Pebblous's guide lists this step as a core safeguard.
- If you bank with Wells Fargo, review the Online Access Agreement before November 24.

## Open questions and what comes next

Several points remain unsettled. Reports disagree on the PNC date, the list of banks differs by outlet, and no regulator had publicly acted as of the coverage. The enforceability of the liability clauses has not been tested in court.

Watch November 24, when Wells Fargo's agreement takes effect and its delegate wording starts to apply to real accounts. Watch whether the CFPB or the OCC says anything about agent-initiated transactions, and whether more banks adopt Section H-style language. The first disputes over an agent-initiated loss will show more about these clauses than any press release, and they will probably start as complaints to a bank before they reach a courtroom.

## Frequently asked questions

### Are banks liable if my AI agent gets hacked?

Under the terms reported so far, the customer generally carries the loss rather than the bank, including when credentials given to an outside agent leak in a breach. Citi's Section H says activity carried out with customer-supplied credentials counts as the customer's own. Whether those clauses survive a dispute under Regulation E is still unsettled, and the reporting shows no court test yet.

### What is Citi's Section H?

Section H is an added section of Citi's online user terms that treats activity carried out with customer-supplied credentials as the customer's own, even if those credentials later leak. It exempts Citi's own first-party AI agents, so the added liability covers only outside tools. Citi may also restrict, block or cut off a third-party agent's access without notice.

### When does Wells Fargo's AI agent rule take effect?

Wells Fargo's Online Access Agreement takes effect November 24, 2026. It treats every transaction by a customer's delegate, including AI agents, as authorized, and Startup Fortune quotes it as covering transactions "including those you did not want or intend." Customers should read the agreement before that date.

### Do I lose money if a third-party AI app is hacked?

The reported terms place the loss on the customer, and banks say they will not reverse agent-initiated transactions, whether mistakes or not. Practical protections include read-only access, transaction alerts, manual approval for transfers and revoking unused agent credentials. Whether a bank must reimburse you anyway depends on how Regulation E applies, and that question remains open.

## Sources

1. [U.S. major banks warn on external AI agents, say customers liable if problems arise](https://www.digitaltoday.co.kr/en/view/127747/us-major-banks-warn-on-external-ai-agents-customers-liable-if-problems-arise)
2. [Citi, PNC and Citizens tell customers they own any loss an AI agent causes](https://startupfortune.com/citi-pnc-and-citizens-tell-customers-they-own-any-loss-an-ai-agent-causes/)
3. [Major US banks crack down on AI agent use by customers](https://www.techradar.com/ai-platforms-assistants/major-us-banks-crack-down-on-ai-agent-use-by-customers-telling-them-theyre-on-their-own-if-it-gets-hacked-or-makes-a-mistake)
4. [Five Major US Banks Disclaim AI Agent Liability, Including After Hacking](https://aigovernance.com/news/five-major-us-banks-disclaim-ai-agent-liability-including-after-hacking)
5. [Citi AI Agent Terms Guide: Liability and Reg E](https://blog.pebblous.ai/blog/citi-ai-agent-liability-terms/en/)
6. [Citi AI Agent Liability Shifts to Customers in New Terms](https://www.tftc.io/citi-ai-agent-liability-customers-section-h)
7. [Citi, PNC Rewrite Terms to Put AI Agent Losses on Customers](https://finance.biggo.com/news/94a3d1e1-8359-4d23-89eb-699326b4a8a2)
8. [AI to ROI News & Analysis: November 20, 2026](https://ai2roi.substack.com/p/ai-to-roi-news-and-analysis-november-78a)
