Third-party agent liability is now a contract term, not a courtesy warning. TechCrunch reported that Citi changed its online terms this week, Citizens emailed a warning, and PNC changed its terms that same day, after JPMorgan and Bank of America announced similar policies last month. Citi still treats outside-agent activity as the customer’s own, including after a credential leak or a hack, while agents the bank provides are carved out.
Updated November 19, 2026
Key Takeaways
- Citi and PNC changed terms this week; Citizens emailed a warning.
- JPMorgan and Bank of America announced similar policies last month.
- Citi still treats leaked or hacked agent use as the customer’s act.
- Bank-provided agents are carved out; Regulation E rights cannot be waived.
What does the first-party carve-out cover?
According to TechCrunch, Citi, the third-largest commercial bank in the United States, updated its Online User Agreement so customers who authorize third-party agents are responsible for activity through those agents. JPMorgan and Bank of America announced similar policies last month. PNC, the eighth-largest bank, changed its terms the day of that report, not on the day of the October announcements, and Citizens emailed a warning TechCrunch saw. DailyAI reports a customer notice that puts Citi’s changes into effect on 24 November 2026.
DailyAI’s reading of the November terms says the new section does not apply to AI tools the bank itself provides through Citi Online or the Citi Mobile app. A Citi spokesperson told TechCrunch the update was meant to “reinforce that customers who choose to grant account access to third-party AI agents are responsible for activities conducted through those agents.”
How far does third-party hack liability reach?
If login details given to the agent are later shared, leaked, or passed on, DailyAI reports that Citi will still treat later use as authorized, with no duty to check even when the action conflicts with the customer’s prompts. Errors and acts beyond the authority granted remain the customer’s, “except as otherwise permitted under applicable law.” The agreement defines an AI agent as software or other technology that takes autonomous, semi-autonomous, or non-autonomous action on behalf of, or at the instruction of, you or any person or entity acting on your behalf.
For ordinary consumer accounts, the contract cannot waive the Electronic Fund Transfer Act or Regulation E. DailyAI, citing Consumer Financial Protection Bureau guidance, notes that a transfer can still be unauthorized when a fraudster uses login details obtained by hacking a third party the customer had shared them with, if the customer did not authorize that transfer and did not benefit from it.
What does third-party agent liability mean for enterprise access?
As analysis, this is the template access owners inherit when a customer’s agent or an employee’s agent touches a system the firm did not build. The owner keeps ordinary treatment for agents it provides, treats a third-party agent as the other party’s act, including after a hack, and may block that agent without notice. A customer’s agent already contacting a service team, or an employee agent on a vendor portal, sits on the exposed side unless the contract says otherwise.
| Access path | What the sources report |
|---|---|
| Agent the bank provides | Citi carve-out: extra liability does not apply |
| Third-party agent the customer authorizes | Citi treats it as the customer’s act, even against the prompt |
| Credentials later leaked or hacked | Citi still treats use as authorized, subject to applicable law |
| Read-only finance tool | Anthropic’s tool “can’t move, manage, or invest your money” |
The WealthAdvisor reported that Jamie Dimon, on JPMorgan’s October earnings call, described the consequences of allowing outside systems into banking relationships as a transfer of liability, while emphasizing that JPMorgan was developing safeguards rather than permitting unrestricted access. Read as analysis, the carve-out is a franchise shield as well as a security clause: the institution keeps the pathway it builds and assigns the rest.
What should access design change before the clause spreads?
Prompt fidelity is not the defense these terms describe. DailyAI reports that Citi has no duty to investigate a clash with the customer’s instructions. Design has to follow the line the clause draws.
- Separate first-party agents from third-party agents in policy, logs, and liability language.
- Do not share usernames and passwords. DailyAI reports Bank of America told customers not to give them to an “AI-enabled tool or agent that offers to act on your behalf.”
- Treat monitoring as yours. DailyAI reports Citi’s privacy notice does not cover how the outside agent stores or shares data.
- Plan for a block without notice — a right DailyAI says Citi reserves — and keep your own stop, such as runtime kill switches on AI agents.
Frequently asked questions
Which banks put third-party AI agent activity on the customer?
TechCrunch reported that Citi changed terms this week, PNC changed its terms that same day, and Citizens emailed customers, after JPMorgan and Bank of America announced similar policies last month. Citi’s version holds even if the tool leaks credentials or is hacked.
Does the extra liability apply to a bank’s own AI tools?
Not under Citi’s update. DailyAI reports the new section does not apply to AI tools the bank itself provides, and TechCrunch says those agents can be used without the additional liability.
Can a bank contract waive unauthorized-transfer rights?
Not for ordinary consumer accounts. DailyAI notes that Regulation E and the Electronic Fund Transfer Act cannot be waived, including when login details were obtained by hacking a third party the customer had shared them with.
What does Citi count as an AI agent?
Citi’s Online User Agreement defines an AI agent as software or other technology that takes autonomous, semi-autonomous, or non-autonomous action on behalf of, or at the instruction of, you or any person or entity acting on your behalf.