Paintings Subscribe Tiếng Việt

22 November 2026

AI News/Banks to customers: let an AI agent into your account and the risk is yours

Citi, Wells Fargo and other US banks just changed their terms: if a third-party AI agent touches your account, everything it does is on you, even after a hack. What this says about the future of agents.

A small piece of news this week matters a lot for anyone building AI agents.

What happened

According to TechCrunch, Citi updated its online banking agreement with a new “Section H”: if a customer lets a third-party AI agent access their account, everything that agent does is the customer’s responsibility, even if credentials are leaked or hacked. Citi’s own agents are excluded. Citi can also block or cut off any outside agent at any time, without notice.

Citi is not alone. According to reports, Wells Fargo updated its agreement, effective 24 November, treating agent-initiated transactions as authorised by the customer. Bank of America, Citizens and PNC have sent similar warnings: do not share your username and password with third parties, AI agents included.

Why it matters

The industry is promising a future where agents pay your bills, move money, book trips and manage spending. But when an agent gets it wrong, or gets hijacked, who is responsible?

The banks’ answer is clear: not them.

I do not think the banks are wrong. They are reacting to today’s reality: most agents access accounts using your own username and password. To the bank, you and the agent are the same person. They cannot tell who pressed the button, so they cannot take responsibility for what they cannot see.

What is missing: identity and permissions for agents

The root problem is not whether agents are smart enough. It is the trust layer:

  • A separate identity for the agent, distinct from the human.
  • Scoped permissions: read balances only, or pay bills below a set amount.
  • An audit trail of every action, who approved it and when.
  • A kill switch to revoke access the moment something looks wrong.

With these in place, banks would have a basis to trust agents, and responsibility could be shared more fairly.

Practical advice for today

  1. Never give your banking password to a third-party AI agent.
  2. If you use a finance app, choose read-only access with no power to move money.
  3. Read your bank’s terms. They differ by bank, and consumer protection law differs by country.

Agents are coming, no doubt. But trust has to be built first, not after.

I wrote more about the limits of moving too fast with agents in Moving faster than we can understand.

Source: TechCrunch