Citi Warns Customers They May Be Liable for AI Banking Agents That Go Rogue

Citigroup has updated its online banking agreement to clarify how responsibility may be handled when customers allow third-party artificial intelligence agents to access their accounts. The revised terms address situations in which an AI tool acts beyond a user's instructions, shares login credentials or enables another party to access the account.

The change highlights a growing challenge for the banking industry: determining who is responsible when autonomous software interacts with financial accounts and makes decisions that affect customers' money.

According to Decrypt, Citi's November agreement treats actions performed using credentials supplied to an external AI agent as authorised by the customer, even when the agent ignores instructions or shares those credentials with another party. The bank says it is not required to investigate whether the agent followed the customer's original prompts, subject to applicable law.

The provision does not apply to AI agents offered directly by Citi.

Banks Are Tightening Rules on Third-Party AI Access

Citi is not the only financial institution reviewing how external AI tools interact with customer accounts.

JPMorgan Chase and Bank of America introduced similar restrictions in October, while Citizens Bank warned customers about the risks of allowing AI assistants to access their accounts. PNC also updated its digital banking agreement.

The changes follow reports of third-party AI agents logging into bank accounts using customer credentials. Some agents reportedly took actions that appeared to go beyond what users had instructed them to do.

Banks face difficulties distinguishing between authorised customer activity and actions performed independently by an external AI system.

Why AI Banking Agents Create New Security Risks

Unlike conventional chatbots, AI agents can perform actions rather than simply provide information. Depending on their permissions, they may navigate banking websites, retrieve financial information and potentially initiate transactions.

This creates several risks:

  • Credential exposure: An AI tool may store or share banking login details insecurely.

  • Unauthorised activity: An agent may perform actions beyond the user's intended instructions.

  • Account access by third parties: Leaked credentials could allow another person or system to access the account.

  • Limited oversight: Customers and banks may struggle to determine exactly what an autonomous agent did and why.

  • Payment risks: Incorrect or unintended financial transactions can create disputes and potential losses.

These concerns are encouraging banks to reconsider how much access external AI systems should receive.

Can Banks Make Customers Responsible for Every AI-Related Loss?

Citi's revised wording does not automatically remove consumer protections under US federal law.

Federal electronic fund transfer rules generally require financial institutions to investigate qualifying reports of unauthorised transactions and limit consumer liability in specified circumstances.

However, the legal position can become complicated when a customer has deliberately shared credentials with an AI agent and the agent subsequently performs an unexpected action.

The distinction between a genuinely unauthorised transfer and a transaction carried out through access granted by the customer may determine how consumer protections apply.

The Consumer Financial Protection Bureau has also stated that banks cannot simply use contractual agreements to waive protections provided under federal law.

As a result, the outcome of an AI-related banking dispute may depend on the circumstances of the incident and the applicable legal requirements.

Regulators and Banks Consider Stronger AI Controls

The US Office of the Comptroller of the Currency has issued examiner guidance addressing risks associated with AI systems and external agents.

The nonbinding framework highlights safeguards such as identity verification, delegated permissions, transaction limits, monitoring and the ability to suspend or revoke access.

PNC's updated agreement also distinguishes between bank-approved connections with limited permissions and unrestricted access provided through shared passwords or one-time security codes.

These approaches indicate that financial institutions may increasingly favour controlled, limited access rather than allowing external AI agents to operate freely inside customer accounts.

What Customers Should Do Before Connecting AI Tools to Bank Accounts

Customers should carefully review the permissions requested by AI-powered financial applications before connecting them to banking services.

Important precautions include:

  • Avoid sharing banking passwords and one-time security codes with third-party AI tools.

  • Use bank-approved account connection methods where available.

  • Review whether an AI application can only read information or also initiate transactions.

  • Enable transaction alerts and monitor account activity.

  • Check the bank's latest digital banking terms and conditions.

  • Revoke access if an application behaves unexpectedly or is no longer needed.

Conclusion

Citi's updated banking terms reflect a wider challenge facing financial institutions as AI agents gain the ability to interact directly with customer accounts.

While autonomous tools may make financial management more convenient, they also introduce questions about access permissions, cybersecurity and responsibility for unexpected transactions.

The future of AI-powered banking will depend on balancing innovation with clear customer protections, controlled access and reliable oversight. As banks develop policies for third-party agents, customers will need to understand exactly what permissions they grant and the risks associated with sharing account access.