InsightsNov 19, 2026 · 5 min read

Banks Just Decided Who Pays When Your AI Agent Gets Hacked. It's Not Them

Citi, PNC, and Citizens just changed the rules on AI agents. If you hand an outside AI tool access to your bank account and that tool gets hacked or leaks your credentials, the loss is yours. Not the bank's.

This follows JPMorgan and Bank of America, who made the same move last month. That puts five banks holding roughly $9.6 trillion in combined assets on the same page: you authorize an AI agent, you own what happens next. Citi's new terms even carve out an exception. If you use Citi's own AI agent instead of a third party one, you don't take on that extra liability. Same convenience, different risk, depending on whose AI you're using.

That last part is the real story. These banks aren't just protecting themselves from fraud. They're drawing a line that steers customers toward their own AI tools and away from everyone else's. Expect the same move from other industries that let AI agents touch sensitive systems: insurers, brokerages, payroll providers, healthcare portals.

Why this matters beyond banking

Most companies are letting employees connect AI tools to internal systems faster than anyone is writing policy for it. A sales rep connects an agent to the CRM. A finance analyst connects one to the ERP. Someone in HR connects one to the payroll system because it saves them two hours a week. Nobody asked IT first.

Banks just showed what happens when that catches up to you. When something goes wrong, the question isn't whether the AI tool made a mistake. It's who authorized it, under what terms, and who absorbs the loss. If your company hasn't answered that question in writing, you're exposed the same way a bank customer with no liability protection is exposed.

What this means department by department

IT and security. You need a clear map of every AI agent with access to company systems, what data it can touch, and what happens if it's compromised. Not a list someone promises to build. An actual inventory, reviewed on a schedule.

Finance. If an AI agent can initiate payments, move data, or touch financial systems, someone needs to define the approval chain before an incident forces the conversation. Waiting for a breach to write the policy is the expensive way to learn this lesson.

Legal and compliance. Vendor contracts for AI tools need the same scrutiny you'd give a payment processor or a data handler. Read the liability language. Know who's holding the bag if something goes sideways.

Every other department. Anyone using an AI agent that touches customer data, financial systems, or credentials needs basic training on what that tool can and can't do, and who's accountable if it's misused.

The takeaway

This isn't really a banking story. It's a governance story that happened to show up in banking first because banks move fast on liability. Large companies rolling out AI agents across departments are going to hit the same question from vendors, insurers, and regulators soon enough: who's responsible when the agent gets it wrong or gets hacked.

The companies that already have an audit of where AI touches their systems, clear approval rules, and trained employees won't be caught flat footed when that question lands on their desk. The ones without it will be writing policy in the middle of a crisis, which is always the worst time to write policy.

If you don't know every AI agent currently touching your company's systems, that's the first gap to close. Not the last one.

RizeTech

Building AI departments inside enterprise companies

Get in touch

Talk about an AI Audit

A short first call, no pitch and no price. We learn how your company runs and where the hours go. If it fits, we scope the audit together. Book a time, or send a note and we reply within one business day.