The honest view from inside technical leadership
Analysis

Engineering leadership

Independent reporting on the decisions, costs and contracts that shape how engineering organizations build with AI.

Agents · News

Citi AI Agent Terms Treat Anything Done With a Leaked Agent Login as Authorized by the Customer

A new Section H in Citi’s online banking agreement makes the customer answerable for a third-party AI agent’s mistakes, and for whatever happens if that agent leaks the password it was handed.

Line-art illustration of a robot AI agent holding a key toward a bank building while dashed copies of the key leak away, representing Citi’s AI agent terms on leaked credentials

Citi’s AI agent terms, added to its online banking agreement in November 2026, say that anything a customer’s third-party AI agent does in their account counts as authorized by the customer. The same rule covers anyone else who gets hold of the login if the agent leaks it, which means the customer carries the loss from a breach at the AI vendor.

Citi’s AI Agent Terms Sit in a New Section H of Its Online User Agreement

The Citi Online User Agreement now opens with a notice that it was last updated in November 2026 to add a Section H titled “Use of Third-Party Artificial Intelligence (‘AI’) Agents.” The section defines an AI agent broadly, as software that acts on a customer’s behalf or instruction whether it runs autonomously, semi-autonomously or not autonomously at all. It then cautions customers against using any third-party agent to give Citi information or instructions or to perform transactions through Citi Online or Citi Mobile. A Citi spokesperson told TechCrunch’s Tim Fernholz that the bank updated the agreement to make the risks of sharing banking credentials with third-party AI tools clearer.

A Leaked Agent Login Counts as the Customer’s Own Authorization

The central clause covers what happens after the customer hands over credentials. Citi says it will consider any access and action by the agent as authorized by the customer. If the credentials given to the agent are then shared, leaked or otherwise passed to other people or other AI agents, Citi will treat their access and actions as authorized by the customer too. The bank also says it has no obligation to check whether such actions are valid, even when they conflict with the prompts or instructions the customer gave the agent. Errors from the agent misreading instructions, unintended actions, and actions beyond the authority the customer granted all fall on the customer, with an exception where applicable law says otherwise. Monitoring the agent is also assigned to the customer in the agreement’s own words.

Citi Can Cut Off Agents at Will, and Its Privacy Notice Stops at the Agent

Section H reserves Citi’s right to limit, suspend or end any third-party agent’s access at any time without notice. It says that by letting an agent into an account, the customer authorizes Citi to disclose their information, including sensitive personal information, to that agent. It adds that the Citi Privacy Notice does not apply to the agent and that Citi does not control how the agent collects, stores, protects or shares the data. The section explicitly excludes AI agents, tools and features that Citi provides itself through its own apps, so the liability shift applies only to outside software.

PNC, Citizens, JPMorgan and Bank of America Have Moved the Same Way

Citi is one of several large US banks drawing the same line. TechCrunch reported that PNC changed its terms on November 18, 2026, and that Citizens emailed customers this week warning they are responsible for what their AI assistant does, based on an email TechCrunch reviewed. JPMorgan and Bank of America announced similar policies in October. TechCrunch also noted that Anthropic’s personal finance tool for Claude, launched earlier in November, is described as read-only and unable to move money.

Why This Matters for Engineering Leadership

Citi’s terms describe the failure mode of password-based delegation in legal language. When an agent logs in as the user, the bank cannot tell the agent from the person, so it treats every action as the person’s, including actions by whoever steals the password from the agent vendor. If your team builds agents that act inside other companies’ systems, credential sharing is now a liability you are handing to your users, and partners can revoke your access without notice. Design for delegated, scoped and revocable authorization from the start: OAuth-style tokens tied to one agent, read-only scopes by default, per-action limits, and an audit trail your users can read. If your company runs the system agents are logging into, Citi’s section is a template for the policy question you will face: decide what an agent identity is, how it differs from a user session, and who carries the loss when one is compromised. Read-only access, as in Anthropic’s finance tool, is the setting that avoids the question entirely.

FAQ

What do Citi’s new AI agent terms say?

Section H of the Citi Online User Agreement cautions customers against using third-party AI agents and says Citi will treat any access or action by such an agent as authorized by the customer.

Is a Citi customer liable if their AI agent leaks their password?

Under the terms, yes. If credentials given to a third-party agent are shared, leaked or distributed further, Citi will consider any access and actions taken with them as authorized by the customer.

Do Citi’s AI agent terms apply to Citi’s own AI tools?

No. The section says it does not apply to any AI agent, tool, function or feature that Citi itself provides through Citi Online or Citi Mobile.

Does the Citi Privacy Notice cover third-party AI agents?

No. The terms say the Citi Privacy Notice does not apply to the agent, and that granting access authorizes Citi to disclose the customer’s information, including sensitive personal information, to it.

Which other banks have changed their terms on AI agents?

TechCrunch reported that PNC changed its terms on November 18, 2026, Citizens emailed customers a similar warning, and JPMorgan and Bank of America announced comparable policies in October.

Source: More banks say customers using AI agents are on their own — even if they get hacked by Tim Fernholz, TechCrunch, November 18, 2026. Primary source: Citi Online User Agreement, Section H, “Use of Third-Party Artificial Intelligence (‘AI’) Agents”, updated November 2026.

Read more news & analysis