What happened

Citibank rewrote part of its online banking agreement this week, and the message to customers is blunt: hand your account login to a third-party AI tool, and whatever happens next is your problem.

Citi is the third-largest commercial bank in the United States. Under the updated terms, customers who let an AI agent into their accounts take responsibility for what the agent does. That covers a bad trade, a wrong bill payment, or a hack. If the tool leaks their credentials, Citi says the customer carries the loss, not the bank.

PNC, the country's eighth-largest bank, changed its own terms on Wednesday. Citizens emailed customers this week with a similar warning. JPMorgan and Bank of America, the two biggest banks in the country, adopted similar policies last month.

A Citi spokesperson told TechCrunch that "customers who choose to grant account access to third-party AI agents are responsible for activities conducted through those agents."

Why it matters

AI agents are moving fast into personal finance. The pitch is simple: give the software your bank login and it will study your spending, suggest investments, pay bills, and move money for you. The banks are now drawing a line under that pitch. They say they will not undo a transaction an automated assistant makes, whether the agent meant to do it or not.

For American customers, the shift matters because it now covers most of the biggest names in US banking. Citi, JPMorgan, Bank of America, PNC, and Citizens together hold accounts for tens of millions of households. Regulators are circling the same question from a different direction: [California has sued OpenAI over incidents involving rogue AI agents](/article/california-sues-openai-rogue-ai-agent-incidents).

The trust gap is real. Plenty of people already hesitate to share an email inbox or a chat history with an AI tool. A bank account raises the stakes. Anthropic, which launched its own personal finance tool earlier this month, built it read-only on purpose. The company says it "can't move, manage, or invest your money."

There is also a business fight underneath the safety language. AI tools that help customers dodge fees or move cash out of low-interest deposit accounts would eat into bank profits, and markets shuddered briefly when that prospect surfaced. Citi's updated terms add one telling detail: AI agents the bank itself provides can be used without the extra liability. The banks want to control the doorway.

The security risk is not theoretical. The FBI says an [AI hacking tool called Artex can breach networks in under three hours](/article/fbi-ai-hacking-tool-artex-breaches-networks-under-three-hours), and [AI models have been tricked into picking civilian targets in war-game tests](/article/ai-models-tricked-picking-civilian-targets-war-game-tests). An agent holding your bank password is a rich target.

What happens next

Expect more banks to follow with similar terms. The pattern is already familiar from other industries: AI labs promise agents will change how people deal with a service, and the service providers push back. Car-hailing and food delivery companies have done the same.

For customers, the practical rule is now clear. Before connecting any AI tool to a bank account, read the bank's terms. A read-only tool that cannot move money carries far less risk than one with full access to your accounts.