Big banks to customers: Use an AI agent, lose your fraud protection

Citibank and major US banks just updated their terms. Give an AI agent your login, and you are fully liable for any hacks or stolen funds.

Business ยท Source: TechCrunch

What happened

Citibank just changed its terms of service to address the rise of automated financial assistants. AI agents from frontier labs and startups can now analyze spending, offer advice, and handle transactions. But if you hand your bank login to a third-party AI agent, you are completely on your own. The bank will not reverse intentional or unintentional transactions made by the bot. More importantly, if the AI tool gets hacked and leaks your credentials, you eat the entire financial loss. The bank considers any action taken through that login as authorized by you.

This is a coordinated industry move across the financial sector. JPMorgan and Bank of America quietly rolled out similar policies last month. PNC and Citizens Bank followed suit this week. They are sending emails and updating online user agreements to make one specific thing clear to consumers. You authorize the AI agent, you authorize the damage. Financial institutions are not thrilled with unreliable AI actors operating inside their systems and making automated decisions.

Banks claim this is strictly about cybersecurity and protecting consumer data. But it is also clearly about market control. Citibank explicitly noted in its terms that customers can use the bank's own provided AI agents without taking on extra liability. They want to own the AI relationship entirely. They are telling customers that third-party tools are dangerous while offering their own automated tools as the safe alternative. This mirrors a broader trend where service providers in food delivery and ride hailing are pushing back against AI labs.

Key facts

Why it matters

Consumer fintech startups building AI agents just hit a massive distribution wall. Breaking into personal finance is an attractive opportunity for AI startups looking to generate revenue from transaction fees. But if using a third-party product voids a user's bank fraud protection, consumer adoption will plummet. People will not risk their life savings just to automate their bill payments. Anthropic already saw this exact problem coming. Their new personal finance tool launched earlier this month is explicitly billed as read-only. It cannot move, manage, or invest your money.

There is a massive second-order effect here regarding bank profitability. Banks are protecting their margins from hyper-competent AI. Markets recently shuddered at the prospect of AI tools making it easier for consumers to avoid unnecessary fees. An AI agent could automatically move money out of low-interest deposit accounts that pad bank margins. By weaponizing liability, banks prevent third-party bots from optimizing consumer finances. This forces startups to rely on official bank connections instead of direct credential access.

For builders

Read only access is the new baseline

Do not ask users for full login credentials. Follow the approach Anthropic took and build read-only financial tools. Consumers will refuse to use your product if it voids their bank guarantees and puts their deposits at risk.

Official bank APIs become gatekeepers

Banks will force startups onto official API connections to maintain control over the ecosystem. Startups will pay the price in restricted access and potential connection fees. Banks win by keeping third-party agents away from their fee-generating products.

Trust is your biggest hurdle

Consumers are already leery of sharing chat records with AI. Asking for bank access is a massive trust challenge. If you build AI agents that handle any sensitive data, your security posture is your main selling point.

My take

I am not surprised the big banks are weaponizing terms of service to kill third-party AI agents. They know a smart agent will instantly move your cash out of their low-yield accounts and block their junk fees. They are using security as an excuse to build a walled garden around your money. Founders need to adapt immediately and stop asking for passwords.

Original reporting: TechCrunch. This is my rewrite and opinion.

More AI news for builders