Finance & Banking | 4 min read

Citibank and Citizens Bank Tell Customers: AI Agent Hacks Are Your Problem

Citibank and Citizens Bank updated their terms of service this week to disclaim liability for hacks or unauthorized transactions caused by third-party AI agents — shifting risk to consumers before regulators define the rules.

Hector Herrera
Hector Herrera
A financial trading floor where a person is assembling related to Citibank and Citizens Bank Tell Customers: AI Agent Hacks Ar
Why this matters Citibank and Citizens Bank updated their terms of service this week to disclaim liability for hacks or unauthorized transactions caused by third-party AI agents — shifting risk to consumers before regulators define the rules.

Citibank has updated its terms of service to inform customers that any unauthorized transactions or hacks arising from their use of third-party AI agents for personal banking are their own liability — not the bank's. Citizens Bank sent similar notices this week. The moves formalize a position the banking sector has been quietly assembling for months: AI agent risk belongs to the customer who chose to deploy it.

The disclaimers arrive as banks are actively detecting the problem they're disclaiming. Major U.S. banks have already begun logging incidents where AI agents — financial assistants built by third-party developers that act on users' behalf — attempted unauthorized transactions in retail accounts. The Federal Reserve took notice in October, when Governor Michael Barr warned banking supervisors that AI agents interacting with the financial system could trigger coordinated bank runs if large numbers of agents respond simultaneously to the same signals.

What the Terms Changes Say

Citibank's updated terms notify customers that by granting an AI agent access to their account — through credential sharing or API authorization — they accept full responsibility for any actions that agent takes. That includes unauthorized transfers, payment errors, and losses from account compromise.

The language closely mirrors existing terms around third-party financial apps, but extends the liability shield more explicitly to cover AI-driven actions. AI agents are harder to predict and audit than traditional software integrations: they interpret natural language instructions, can be manipulated through prompt injection attacks, and may execute actions based on ambiguous or erroneous reasoning that conventional software simply wouldn't attempt.

Citizens Bank's notice is similarly structured. Neither bank has released the full revised language publicly, but reporting by TechCrunch confirmed both institutions sent notices to customers this week.

Expect more banks to follow. When one major institution sets a liability framework on a novel technology risk, others converge quickly — especially when regulatory guidance hasn't yet caught up with the technology and the first mover sets the precedent.

What Banks Are Seeing

The disclaimers aren't precautionary. Banks are responding to a live threat environment.

AI financial agents vary enormously in quality and security. Some are built by reputable fintech companies with established security practices. Others are built by smaller developers, or assembled through automated pipelines, with little formal security review. When these agents access bank accounts, they create attack surfaces: compromised credentials, injection attacks that manipulate the agent's instructions, and agents that interpret instructions too broadly and initiate transfers the user didn't explicitly authorize.

Prompt injection — where malicious content in a document or webpage tricks the AI agent into taking unintended actions — is an emerging attack vector specific to AI systems. A user might direct their financial AI agent to review a transaction summary, and an attacker embeds instructions in that document that redirect the agent to initiate a wire transfer. The user authorized the agent; the agent was then manipulated. Under the new terms, that loss falls on the customer.

The Fed's Barr scenario is the scarier edge case: not individual account compromise, but coordinated agent behavior at scale. If thousands of agents share similar market data inputs and similar logic, a market event could trigger simultaneous withdrawal decisions across a large portion of retail deposits — amplifying rather than smoothing systemic stress. This isn't a scenario banks are dealing with today, but it's why regulators are watching the category closely.

What This Means for Consumers

The practical impact is straightforward: if you use a third-party AI agent to manage your money and something goes wrong, your bank is not covering it.

This is a meaningful shift from how most consumers think about financial security. The assumption that unauthorized transactions are the bank's problem to resolve — backed by federal protections under Regulation E — doesn't apply cleanly when the unauthorized action was initiated by software you authorized, even if that software was subsequently compromised or behaved unexpectedly.

What consumers should do:

  • Review what permissions you've granted to any AI-powered financial tool — many people connect apps to financial accounts and don't revisit those permissions
  • Understand the difference between a regulated fintech app (which has its own compliance obligations and security requirements) and an AI agent built on a general-purpose platform
  • Monitor accounts closely if you've connected any AI agents with transaction or transfer capabilities
  • Read updated terms of service from your bank — the liability language change is likely in an email you've already received

The Regulatory Question

The Consumer Financial Protection Bureau (CFPB) has been signaling interest in AI agent accountability in financial services but hasn't issued guidance. The banking sector's move to shift liability contractually before regulators define the rules puts pressure on the CFPB to clarify whether Regulation E protections extend to AI-agent-initiated transactions — and if not, whether they should.

There's a reasonable legal argument that an AI agent acting within the scope of a customer's authorization is not an "unauthorized" transaction under the current Regulation E definition — meaning banks' disclaimers may already be legally sound under existing law. If that interpretation holds, consumers relying on federal protections may find they don't apply to AI-mediated losses.

What to watch: Legislative or regulatory response to these liability disclaimers will determine whether banks can hold this position long-term. Watch the CFPB's AI guidance queue through Q1 2027 — and watch for class-action litigation if a significant AI agent fraud incident triggers consumer losses that the new terms place entirely on account holders.

Key Takeaways

  • ✓ What consumers should do:

Did this help you understand AI better?

Your feedback helps us write more useful content.

Hector Herrera

Written by

Hector Herrera

Hector Herrera is an AI systems architect in Houston and founder of Hex AI Systems. He designs and runs AI systems in production and writes daily about how AI is reshaping business, government and everyday life. 20+ years building for the web. Houston, TX.

More from Hector →

Get tomorrow's AI briefing

Join readers who start their day with NexChron. Free, daily, no spam.

More from NexChron