Articles · Q&A
Should you give an AI app access to your bank account? Who pays if it goes wrong, explained for India
By AI Week Daily editors · Published 19 November 2026 · 5 min read
AI assistants can now read bank statements, track spending and, in some cases, make payments for you. This week Citibank changed its terms to say that customers who give third-party AI agents access to their accounts are responsible for whatever those agents do, even if the AI tool is hacked, and JPMorgan, Bank of America, PNC and Citizens have taken similar positions (TechCrunch). Those are US banks, but the question applies in India too: if an AI app with your banking access makes a wrong payment, or leaks your login, who pays? Here are the answers, based on the Reserve Bank of India's rules on unauthorised electronic transactions.
What exactly did the US banks change?
Citi updated its online user agreement to say customers who grant account access to third-party AI agents are responsible for activities conducted through those agents. It goes further: if the AI tool leaks your credentials or is hacked, you are responsible for what the hackers do with your account. Citi said its own AI agents can be used without that extra liability (TechCrunch).
Banks are also building agents themselves. JPMorgan and Hyundai Card recently completed a proof-of-concept in which AI agents handled a simulated foreign-exchange transaction, and are aiming for a more complex phase in the first half of 2027 (CNBC). So the rules about who carries the risk are being written now.
Does any of this apply to Indian bank accounts?
Not directly. Indian banks follow RBI's directions on customer protection in unauthorised electronic banking transactions, which every scheduled commercial bank, small finance bank and payments bank must turn into a board-approved policy and publish (RBI). Those directions do not mention AI agents. What matters is how your situation fits the categories RBI does define.
When does RBI say you lose nothing?
RBI gives the customer zero liability in two cases (RBI):
- The fraud happened because of negligence or a deficiency on the bank's side, whether or not you reported it.
- The breach was by a third party, the fault lies neither with you nor the bank but elsewhere in the system, and you told the bank within three working days of receiving its alert about the transaction.
Once you report, the bank must credit the disputed amount to your account (a 'shadow reversal') within 10 working days, without waiting for any insurance claim. The burden of proving that you are liable lies on the bank (RBI).
What if I report late?
For a third-party breach reported within four to seven working days, your liability per transaction is capped at the transaction value or these limits, whichever is lower (RBI):
- ₹5,000 for basic savings bank deposit (BSBD) accounts.
- ₹10,000 for other savings accounts, prepaid instruments and gift cards, many individual current/overdraft accounts, and credit cards with limits up to ₹5 lakh.
- ₹25,000 for other current/overdraft accounts and credit cards with limits above ₹5 lakh.
Beyond seven working days, liability is decided by your bank's own board-approved policy (RBI).
Here is the catch: what counts as 'sharing credentials'?
RBI says that where the loss is due to the customer's negligence, "such as where he has shared the payment credentials", the customer bears the entire loss until the unauthorised transaction is reported (RBI). Typing your net-banking password, card PIN, UPI PIN or an OTP into an AI app, or letting it log in as you, can look very much like sharing credentials. A bank could argue that a payment made by an agent you set up was not 'unauthorised' at all. RBI's text does not settle that, so do not assume the zero-liability rule will protect you.
A worked example
Suppose an AI budgeting app you installed leaks your net-banking login and ₹40,000 is transferred out of your savings account overnight. You see the bank's SMS alert at 7 am.
- You report the same morning through the bank's helpline or by replying to the alert, and keep the complaint number. If the bank accepts this as a third-party breach, you have zero liability and the ₹40,000 should be shadow-credited within 10 working days.
- You report on day five. If it is treated as a third-party breach, your loss is capped at ₹10,000 for a regular savings account.
- The bank says you shared your credentials with the app. Then you may bear the loss up to the moment you reported. Any money taken after you reported is the bank's responsibility.
These outcomes follow RBI's categories (RBI); which one applies to you is decided by your bank, and disputes can be escalated through its grievance process.
How to use AI for money without handing over the keys
- Prefer read-only. Download your statement as a PDF or CSV and upload it, or paste in figures, instead of giving an app your login. Anthropic's new personal finance tool, for example, is billed as "read-only" and says it "can't move, manage, or invest your money" (TechCrunch).
- Mask what you upload. Remove account numbers and addresses from statements before sharing them with any chatbot.
- Never share a UPI PIN, card PIN, CVV or OTP with any app, bot or person, and do not let an AI tool see OTP messages.
- Keep SMS alerts on and act on any transaction you do not recognise the same day. RBI requires banks to offer 24x7 reporting channels, including replying to the alert (RBI).
- Read your bank's customer-protection policy, usually on its website, for how it treats third-party apps.
The bottom line
AI tools are genuinely useful for spotting wasteful spending and planning budgets, and they can do most of that from a statement alone. Give an app the power to move money only when your bank offers it directly and states in writing who carries the risk. Until then, the cheapest protection is to keep your passwords, PINs and OTPs to yourself and report anything odd within three working days.
Sources
- TechCrunch: More banks say customers using AI agents are on their own — even if they get hacked (18 November 2026)
- Reserve Bank of India: Customer Protection – Limiting Liability of Customers in Unauthorised Electronic Banking Transactions (RBI/2017-18/15, 6 July 2017)
- CNBC: JPMorgan is working on AI agent transactions that don't involve humans (18 November 2026)
Original article by AI Week Daily editors. Facts and figures are as stated by the linked sources on the publish date; check them for updates. Spotted an error? Tell us.
More articles
- How to find and cancel forgotten subscriptions with an AI assistant: a step-by-step guide for India
- Before you let an AI agent shop for you: a 9-point money checklist
- Can your phone's AI assistant read your banking app? What Signal's screenshot block reveals, and the settings to check
- Anthropic's IPO from India: a Q&A on access, TCS, tax and the first-day trap
- Will AI cut your pay or raise it? What three big studies actually found, and what to do with your career money