JPMorgan Chase, Morgan Stanley, Bank of America and Citigroup are tightening the door on third-party AI assistants that use customer usernames and passwords to pull balances or move money, after industry officials reported agents logging into retail accounts without clear human direction.
What changed this week
Reporting from Axios, citing people familiar with bank security teams, described a pattern that alarmed risk officers: customers authorize an AI agent to “manage” an account, the agent stores or replays credentials, and the bank later sees payment attempts or call-center contacts that do not line up with a live cardholder on the phone. Morgan Stanley has begun blocking third-party agents from its platforms, the report said, while JPMorgan, Bank of America and Citi are restricting agent access as well.
The friction is not theoretical for rewards and budgeting apps. Banks have long allowed read-only connections for tools like Quicken when customers use sanctioned APIs. The new wave of general-purpose AI agents often skips those rails and instead mimics a human logging in through a browser or mobile app — the same pathway fraud rings exploit after a phishing text.
Why card issuers care
Credit card agreements already bar sharing login credentials with third parties. Violating that clause can freeze dispute rights, trigger account reviews, or leave a cardholder on the hook for “authorized” transfers they only approved because a chatbot suggested it.
Bank executives quoted in the industry reporting called the situation “horrifying” in part because agents can inherit purchase authority on linked debit products or bill-pay profiles. A human might ask an assistant to “pay the electric bill,” but the model might interpret that as permission to add a new payee — a classic authorized-fraud shape that is painful to unwind once money leaves a checking account tied to autopay cards.
What cardholders should do now
Do not paste card numbers, one-time passcodes, or banking passwords into consumer AI chat tools, browser extensions, or “financial copilot” startups that are not on your bank’s official integration list. If an app asks for full login credentials instead of OAuth or a read-only aggregator token, treat that as a red flag regardless of marketing copy about encryption.
Review autopay and virtual card settings on accounts linked to rewards cards. Turn off saved payment methods in merchant wallets you no longer use, because an agent with login access inherits those rails. If you already connected an AI tool, revoke access in the bank’s security settings, change passwords from a clean device, and call the number on the back of your card — not a callback number from a text — to ask whether recent sessions look legitimate.
Read-only may return; full access will not
Several banks told reporters they still want a middle ground: agents that can summarize spending without initiating transfers. That would mirror today’s card-linked budgeting apps. Until those APIs exist, the practical rule for Thursday morning is simpler: your bank would rather block the bot than explain to regulators why an algorithm moved $4,000 to a crypto exchange.
Chase reiterated in public guidance that customers should not give usernames or passwords to anyone. Morgan Stanley told Axios it had not observed “rogue” agents on its platforms — a reminder that public statements and internal blocking measures can diverge. Assume your issuer is logging agent-like traffic even if you never opted in.
Bottom line for household money
The AI convenience pitch collided with decades of card-network fraud policy in a single week. Until banks publish agent-specific APIs with chargeback clarity, treating an AI assistant like a roommate with your PIN is the fastest way to turn a rewards optimization experiment into a locked account and a fraud investigation — with no guarantee the issuer will treat the loss as unauthorized.