Three large U.S. banks have now told customers that they, not the bank, will carry the risk when a third-party AI agent acts inside their accounts, even if the agent is hacked or its login is stolen. Citi, PNC and Citizens made their moves during the week of November 18, 2026, following JPMorgan Chase and Bank of America, which announced similar policies in October. The timing matters because consumer agents that can log in and move money are spreading faster than the rules that govern them.

What Each Bank Changed

Citi made the most sweeping change. It added a new Section H to its Online User Agreement during the week of November 18, stating that customers who give third-party agents access are responsible for the activity conducted through them. PNC changed its terms on November 18. Citizens took a lighter route, emailing customers to warn that they would be responsible for anything their AI assistant does. Reporting on Citizens did not give a more precise date than that week.

The Citi language is the most detailed of the three, and several provisions stand out:

  • Scope: Section H covers anyone who lets a third-party AI agent access accounts or digital services through Citi Online or Citi Mobile.
  • Mistakes count: Citi treats whatever an agent does as authorized by the customer, including errors and actions beyond the authority the customer intended.
  • Transfers: Under a separate funds transfer agreement, a transfer started by an agent counts as authorized, even when it conflicts with the customer's instructions.
  • Stolen logins: If credentials are leaked or stolen and used by someone else or by an agent, Citi says it will still treat those actions as the customer's and has no obligation to investigate whether they were valid.
  • Monitoring and access: Watching what an agent does is the customer's job, not Citi's, and Citi can cut off agent access at any time without warning.

The Bank-Built Exception

Citi's terms exempt AI tools that it builds and offers itself through Citi Online or Citi Mobile. That distinction reveals where the bank places the risk. When an agent runs inside the bank's own systems, the bank accepts responsibility. When a customer hands login access to outside software, the customer carries the risk, regardless of how that outside tool behaves.

For consumers, this makes the choice of tool more consequential than it may appear. An agent that logs into a bank account on a user's behalf is not just a convenience. Under these terms, it is a source of legal exposure that the user may not be able to pass back to the bank.

Why It Matters

The larger issue is that these agreements are being tested in the market before courts have weighed in. The legal enforceability of such liability shifts is unsettled, and it is not yet clear how far a bank can push responsibility for losses caused by a breach it did not cause. Regulators and consumer advocates are likely to watch whether these clauses hold up when a stolen credential leads to a large unauthorized transfer.

There is also a privacy dimension. Citi's terms say that using a third-party agent means agreeing to let the bank share personal information with it, and that Citi's privacy notice does not cover what the agent does with that data. The bank says it cannot control how the agent handles it. For users, that is a second layer of risk on top of the question of who pays after a loss.

The move also reflects a wider problem for the financial sector. Banks have built fraud controls around human logins, passwords and one-time codes. Agents that operate continuously, switch between apps and carry out multistep actions do not fit those controls neatly. The banks appear to be choosing to shift the burden to customers rather than rebuild controls for third-party access.

What Account Holders Can Do Now

Customers who already let outside apps or assistants touch their accounts have several practical steps to take:

  • Review which third-party apps hold account credentials or connected access, and remove any you no longer use.
  • Prefer agents that connect through the bank's own official tools, which Citi exempts from the new liability language.
  • Read any updated user agreement and any notices about funds transfers, since transfers started by an agent may be treated as authorized by default.
  • Limit what an agent can do, such as using read-only access where the service offers it.

The new clauses are a reminder that convenience and liability now arrive bundled together. Before handing an AI agent access to a bank account, customers should understand that under the latest terms at several large banks, they may be the ones left holding the bill.

Sources