---
title: "Citi Warns Customers Against Third-Party AI Agents"
description: "Citi's Section H makes customers liable for third-party AI agent actions, even with shared or leaked credentials. Key lessons for brand operations teams."
canonical: https://epinium.com/en/blog/citi-warns-customers-against-third-party-ai-agents/
lang: en
date: 2026-11-22T06:05:44
---

**Executive summary**

- **What happened:** Citi added a new Section H to its Citi Online User Agreement, titled "Use of Third-Party Artificial Intelligence ('AI') Agents", warning customers off third-party agents and making them responsible for what those agents do.
- **Impact:** Citi treats every access and action by the agent as authorized by the customer, even if credentials are shared, leaked or passed to other people or AI agents.
- **What matters for brands:** Citi's own AI tools are excluded from Section H. The rule is "our agents yes, yours at your risk", and your commerce channels will probably follow the same logic.

Picture the scene. Your finance lead connects a clever AI agent to the company bank portal so it can reconcile payouts. A week later something moves that nobody approved. Who pays?

According to Citi's new terms, you do.

## What Citi actually changed

In November 2026, Citi added Section H to its Citi Online User Agreement. It applies when third-party AI agents reach digital accounts or services through Citi Online or Citi Mobile, per [Unite.AI's coverage](https://www.unite.ai/citi-makes-customers-responsible-for-third-party-ai-agent-activity/). [Dig.watch reports](https://dig.watch/updates/citibank-update-user-agreement-on-ai-liability) that Citibank revised the agreement on 18 November 2026. The section is titled "Use of Third-Party Artificial Intelligence ('AI') Agents", according to [TFTC](https://www.tftc.io/citi-ai-agent-liability-customers-section-h).

The content is blunt. Citi warns customers against using third-party agents to access their accounts, give instructions to the bank or make transactions. It reserves the right to limit, block, suspend or terminate any third-party agent's access at any time and without notice. And it will treat all access and actions by the agent as authorized by the customer, even if credentials are shared, leaked or distributed to other people or AI agents, as [Towards AI](https://pub.towardsai.net/citi-just-made-you-liable-for-your-ai-agent-write-the-mandate-before-you-plug-it-in-552843b8f260?gi=1cce77d77fde) describes.

There's a second document too. The Citibank Agreement for Online Funds Transfers also got a November 2026 update, and its Section 4 makes the customer solely responsible for instructions the agent sends to Citi ([Unite.AI](https://www.unite.ai/citi-makes-customers-responsible-for-third-party-ai-agent-activity/)).

A Citi spokesperson told [TechCrunch](https://techcrunch.com/2026/11/18/more-banks-says-customers-using-ai-agents-are-on-their-own-even-if-they-get-hacked/) that customers who grant access to third-party AI agents are responsible for the activity carried out through them.

One detail deserves more attention than it's getting: agents, tools or functions provided by Citi itself are excluded from Section H ([TFTC](https://www.tftc.io/citi-ai-agent-liability-customers-section-h)).

## This isn't a banking story, it's a delegation story

Most readers will file this under fintech. I'd file it under operations.

Every brand and manufacturer is now deciding which agents get which keys: to the ERP, the marketplace dashboards, the ad accounts, the storefront. Citi has just written down, in plain contract language, the default answer platforms will give when something goes wrong: you delegated, you own it.

We've seen the same direction elsewhere. Amazon is [positioning Alexa for shopping to block third-party AI agents](/en/blog/amazon-positions-alexa-for-shopping-to-block-third-party-ai-agents/), while Shopify [enables AI agent checkouts](/en/blog/shopify-enables-ai-agents-checkouts/) on its own terms. Platforms are not neutral. They're choosing which agents are welcome, and their terms decide who carries the liability.

Now the contrarian take. The popular reading is "Citi is being paranoid and anti-innovation." I think it's the opposite: Citi is the first to say out loud what every platform legal team already believes. A clause that looks hostile is actually a spec for how to deploy agents responsibly. Complaining about it is a waste of energy. Reading it as a requirements document is useful.

Is the clause the last word? Not necessarily. [Decrypt notes](https://decrypt.co/383677/citi-customers-responsible-ai-banking-agents-rogue-hack) that federal consumer fraud protections can't simply be removed by a bank agreement, although how they apply to transactions by customer-authorized AI agents is less clear. Less clear is not protected. Don't build your risk policy on a legal gray area.

## What brand and operations leaders should do this week

Treat agent access like you treat employee access, because contractually that's how it will be treated.

- Inventory every agent, plugin and automation that holds credentials to a financial, commerce or ad system.
- Give each agent its own scoped credentials. Never share a human login, since shared or leaked credentials count as your authorization.
- Separate read from write. Reading a dashboard is a very different risk from moving money or editing listings.
- Assume access can be cut without notice. Any workflow that depends on a third-party agent needs a manual fallback.
- Ask your vendors one question: if your agent errs, what does your contract say?

This is also where teams that don't know where to start with AI get stuck. The problem is rarely the model. It's the missing mandate: who can the agent act for, on what, up to what limit. If you want the technical side of that, our piece on [why enterprise AI agents fail without an agentic context layer](/en/blog/why-enterprise-ai-agents-fail-agentic-context-layer/) covers it, and [why AI agents abandon retail shopping carts](/en/blog/why-ai-agents-abandon-retail-shopping-carts/) shows what friction looks like from the agent's side.

> **Section H** — Citi will treat all agent access and actions as authorized by the customer, even if credentials are shared, leaked or distributed. [Source: Towards AI](https://pub.towardsai.net/citi-just-made-you-liable-for-your-ai-agent-write-the-mandate-before-you-plug-it-in-552843b8f260?gi=1cce77d77fde)

FREE SESSION
**Not sure which agents should touch your stack?** Epinium is the AI technology partner for brands and manufacturers: we build the technology you're missing and operate it with our platform. [See Epinium’s services →](https://epinium.com/en/services/)
free 15-min diagnostic

## FAQ

### What did Citi change in its user agreement?
Citi added Section H to the Citi Online User Agreement, titled "Use of Third-Party Artificial Intelligence ('AI') Agents". It applies when third-party agents access accounts or services via Citi Online or Citi Mobile, and it warns customers against using them (source: [Unite.AI](https://www.unite.ai/citi-makes-customers-responsible-for-third-party-ai-agent-activity/)).

### Who is liable if a third-party AI agent makes a mistake or is hacked?
Under Citi's terms, the customer. A Citi spokesperson told [TechCrunch](https://techcrunch.com/2026/11/18/more-banks-says-customers-using-ai-agents-are-on-their-own-even-if-they-get-hacked/) that customers granting access are responsible for the activity carried out through those agents.

### Does this apply to Citi's own AI tools?
No. Agents, tools or functions provided by Citi itself are excluded from Section H ([TFTC](https://www.tftc.io/citi-ai-agent-liability-customers-section-h)).

### Can Citi block an agent?
Yes. It reserves the right to limit, prevent, suspend or terminate any third-party agent's access at any time and without prior notice ([Unite.AI](https://www.unite.ai/citi-makes-customers-responsible-for-third-party-ai-agent-activity/)).

### Does this affect brands that don't bank with Citi?
Directly, no. Strategically, yes: it shows how platforms may allocate liability for delegated agent actions. Review your agent permissions, scoped credentials and fallbacks now, whatever systems you connect.

## The takeaway

Citi didn't ban agents. It priced the risk and handed it to the customer. Your move is to decide, in writing, what each agent may do before a platform decides for you.

SERVICES BY EPINIUM
**Put a mandate on every agent before it goes live.** Epinium has more than 10 years in retail, and its Services line helps brands and manufacturers set up AI integrations and automations safely. [Book free diagnostic →](https://epinium.com/en/contact/)
free 15-min diagnostic

<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@graph": [
    {
      "@type": "FAQPage",
      "mainEntity": [
        {
          "@type": "Question",
          "name": "What did Citi change in its user agreement?",
          "acceptedAnswer": {
            "@type": "Answer",
            "text": "Citi added Section H to the Citi Online User Agreement, titled \"Use of Third-Party Artificial Intelligence ('AI') Agents\". It applies when third-party agents access accounts or services via Citi Online or Citi Mobile, and it warns customers against using them (source: Unite.AI)."
          }
        },
        {
          "@type": "Question",
          "name": "Who is liable if a third-party AI agent makes a mistake or is hacked?",
          "acceptedAnswer": {
            "@type": "Answer",
            "text": "Under Citi's terms, the customer. A Citi spokesperson told TechCrunch that customers granting access are responsible for the activity carried out through those agents."
          }
        },
        {
          "@type": "Question",
          "name": "Does this apply to Citi's own AI tools?",
          "acceptedAnswer": {
            "@type": "Answer",
            "text": "No. Agents, tools or functions provided by Citi itself are excluded from Section H (TFTC)."
          }
        },
        {
          "@type": "Question",
          "name": "Can Citi block an agent?",
          "acceptedAnswer": {
            "@type": "Answer",
            "text": "Yes. It reserves the right to limit, prevent, suspend or terminate any third-party agent's access at any time and without prior notice (Unite.AI)."
          }
        },
        {
          "@type": "Question",
          "name": "Does this affect brands that don't bank with Citi?",
          "acceptedAnswer": {
            "@type": "Answer",
            "text": "Directly, no. Strategically, yes: it shows how platforms may allocate liability for delegated agent actions. Review your agent permissions, scoped credentials and fallbacks now, whatever systems you connect."
          }
        }
      ]
    },
    {
      "@type": "Person",
      "name": "Epinium Editorial Team",
      "worksFor": {
        "@type": "Organization",
        "name": "Epinium",
        "url": "https://epinium.com"
      }
    }
  ]
}
</script>