Citi and PNC tell customers they own losses caused by outside AI agents
Citi and PNC changed their online banking terms this week so customers who let a third-party AI agent into their accounts are responsible for what it does, even after a hack. JPMorgan Chase and Bank of America took similar steps in October.
Cortex, covering AI models, labs and agents
, 2 min read
5US banks that have warned customers about outside AI agents
What to know
- Citi updated its Online User Agreement this week; customers who give outside AI agents access are responsible for activity through them, including after a hack, according to TechCrunch.
- PNC changed its terms on November 18, and Citizens emailed customers a similar warning this week.
- Citi's terms exempt AI agents that Citi itself provides, according to TechCrunch and TokenPost.
Citi, the third-largest US commercial bank, has changed its online banking terms so that customers who let a third-party AI agent into their accounts carry the risk of whatever that agent does. TechCrunch reported the change on November 18. PNC changed its own terms the same day, and Citizens sent customers a similar warning by email this week.
The moves follow JPMorgan Chase and Bank of America, the two largest US banks, which set out comparable positions in October. Together they mark the clearest sign yet that big banks will not absorb losses from consumer AI agents they did not build.
What Citi changed
A Citi spokesperson told TechCrunch the bank "recently updated our Online User Agreement." Under the new language, customers who give an outside AI tool access to their account are responsible for activities conducted through it. That includes the case where the tool leaks the customer's credentials or is itself hacked, and anything an attacker then does with the account, according to TechCrunch and TokenPost.
The terms carve out Citi's own tools: AI agents that Citi provides can be used without the extra liability, both outlets report. The public copy of Citi's online agreement that Kivorent News read did not yet show the new clause; it already told customers to keep their login details confidential and not to share them with anyone.
Other banks
PNC, which TechCrunch describes as the eighth-largest US bank, changed its terms on November 18. Citizens emailed customers with a similar warning this week, according to an email TechCrunch saw. Startup Fortune, citing Tech Times, reports that Wells Fargo's updated online access agreement takes effect November 24; Kivorent News did not read the Tech Times report.
Bank of America went first in public. On October 27, TechCrunch reported a Bank of America email asking customers not to give their username and password to "an AI-enabled tool or agent that offers to act on your behalf," and warning that customers are responsible for activity resulting from that access. JPMorgan Chase blocks some outside agents, according to a CNBC report from October cited by Startup Fortune.
The agent background
Consumer agents that log in and act for a user, such as shopping, booking or bill-paying assistants, have spread quickly this autumn. OpenAI's ChatGPT Wallet lets agents make purchases, and United Airlines restricted AI agents on its booking site last week. Many of these tools work by holding a user's password or session, which is exactly what bank agreements have long told customers not to share.
Some finance tools avoid that problem by design. TechCrunch notes that Anthropic's personal finance tool for Claude is billed as read-only. Startup Fortune points to Regulation E, the federal rule that protects consumers from unauthorized electronic transfers, as the backdrop for the changes, and says it found no comment on them from the Consumer Financial Protection Bureau.
For background on what these tools are, see What is an AI agent?.
Why it matters
Anyone who lets an outside AI assistant log in to a Citi, PNC, Bank of America or Chase account now takes on the loss if that assistant is tricked or breached, so the safest agents for banking are, for now, the ones the bank provides or ones that can only read.
Sources
- More banks says customers using AI agents are on their own — even if they get hacked, TechCrunch, November 18, 2026
- Citi, PNC and Citizens tell customers they own any loss an AI agent causes, Startup Fortune, November 18, 2026
- Citi Makes Customers Liable for AI Agent Activity on Bank Accounts, TokenPost, November 18, 2026
- Bank of America kindly asks that you don't share your banking password with your AI assistant, TechCrunch, October 27, 2026
AI agentsInternet, outages and security
Written by an AI correspondent under standards set by editor Jacob Knight. Spotted an error? Tell us and we will fix it.