Guide
AI Bots Are Inside Bank Accounts. Is Your Money Safe?
The Fed says AI agents are already inside customer bank accounts — and they could move money faster than any bank run in history. Here's how to protect yours.

In this article
- What the Fed actually warned about
- What is an AI agent — and how does it get into your account?
- What the banks are already doing
- Why speed changes everything: the Silicon Valley Bank lesson
- What your money is actually protected by
- Seven steps to keep your money under your control
- 1. Never give an AI agent your banking password
- 2. Audit what is already connected to your accounts
- 3. Switch on passkeys or biometric login
- 4. Turn on transaction alerts for everything
- 5. Keep your spending money separate from your savings
- 6. Read your bank's updated terms before 2026 ends
- 7. Know exactly what to do if a transfer looks wrong
- Frequently asked questions
- Key takeaways
On 16 November 2026, Federal Reserve Governor Michael Barr told an AI finance conference in Milan something no central banker had put on the record before: major banks are already finding AI agents inside ordinary customers' bank accounts, trying to move money, without the bank being able to tell whether the customer asked them to. The same software could one day empty a bank in minutes. Here is what he said, what banks are doing about it, and seven steps to keep your money under your control.
What the Fed actually warned about
Barr's speech, "Some Potential Financial Stability Implications of Artificial Intelligence", was delivered at the 7th ACM International Conference on AI in Finance in Milan — his fourth speech in a year devoted entirely to AI risk. It came three days after the Fed's Financial Stability Report named AI the most cited shared risk to financial stability, with more than 85% of those surveyed listing it as a risk factor.
The passage that matters most to anyone with a bank account is short. Barr said:
"Major banks now report finding AI agents attempting to take actions in retail customer accounts, without the bank being able to verify authenticity, consent and alignment with customer instructions."
This is not a warning about a future technology. Banks are already detecting agents operating inside customer accounts, and in some cases they cannot verify three basic things: that the agent is who it claims to be, that the customer consented, and that its actions match what the customer actually asked for.
Barr's second warning is about speed. "AI agents could instantaneously detect yield differentials or stability signals and autonomously move funds at high speed," he said, "potentially accelerating runs that unfold in minutes, overwhelming traditional liquidity management and regulatory intervention timelines." A bank run used to take days; if thousands of agents spot the same warning sign at once, it could take minutes.
What is an AI agent — and how does it get into your account?
An AI agent is software that takes a goal — "find me the best savings rate", "pay this bill", "move my money somewhere safer" — and carries out the steps itself: browsing, logging in, clicking, and transferring, without you approving each move. The tools are already mainstream: ChatGPT's agent mode can operate a browser on your behalf, Coinbase has just launched a plugin for trading inside a chat window, and Visa is building the rails for agents to shop with your card.
The problem is the gap between what an agent can do and what anyone can verify. Axios reported on 22 October 2026 that large US banks were already finding third-party AI agents inside customer accounts, using logins customers had handed over — and sometimes acting beyond what the customer asked. The agents were not necessarily hacking anything; in many cases the customer gave them the keys, and the agent then did more than intended.
The distinction that matters: an AI tool that advises you costs nothing if it is wrong; one that acts for you can move real money before you notice. Before connecting any AI tool to a financial account, check whether it needs your password — and whether it can transact, not just read.
What the banks are already doing
The banks are not waiting for regulators. On its third-quarter 2026 earnings call, JPMorgan Chase chief executive Jamie Dimon said the bank was "not going to allow just any outside agent to come in inside the company because that's just too risky for clients and for JPMorgan." American Banker reported on 27 October that at least five of the largest US banks — Bank of America, Citi, JPMorgan Chase, Morgan Stanley and Wells Fargo — are blocking or restricting outside AI agents.
Wells Fargo has gone further in writing: its updated online access agreement, effective 24 November 2026, says that unless the law provides otherwise, an authorised agent's actions are treated as the customer's own. If you authorise an agent and it does something you did not want, the bank's starting position may be that you did it.
AI is also being used to attack: CrowdStrike's 2026 Global Threat Report found exploits from "AI-enabled adversaries" jumped 89%, and Barr warned that "attackers only need to be right once, while defenders need to be right all the time." The same technology also defends — Visa recently noted a 35% rise in criminals probing for valid payment accounts, while its systems blocked 200% more bogus transactions than in the prior six months, largely thanks to AI.
Why speed changes everything: the Silicon Valley Bank lesson
To understand why a Fed governor worries about minutes, look at the last big bank run. In March 2023, Silicon Valley Bank lost more than $40 billion of deposits in a single day; over two days, customers tried to withdraw about $142 billion — 81% of its $175 billion in deposits. Barr led the Fed's own review of that failure. That run was driven by humans on smartphones. Now imagine the same panic with software doing the spotting, deciding and transferring — and thousands of agents built on similar models all reaching the same conclusion at once, possibly tricked by spoofing, the fake signals Barr warned "may echo unfortunate human tendencies".
No AI-driven bank run has happened, and Barr's language was conditional throughout — "could", "may", "potentially". The speech is one governor's view, not a change in Federal Reserve policy. But the agents inside accounts are real, the banks blocking them are real, and the speed problem is arithmetic, not speculation.
What your money is actually protected by
Deposit insurance has not changed. What changes is how fast a problem could arrive — and insurance protects the amount, not the timing.
| Protection | United States | United Kingdom |
|---|---|---|
| Deposit protection limit | FDIC insurance: $250,000 per depositor, per insured bank, per ownership category | FSCS protection: £85,000 per person, per banking licence (banks sharing a licence count as one) |
| How fast you get it back if a bank fails | FDIC typically makes insured deposits available within a few business days — after Silicon Valley Bank failed on a Friday, insured depositors had access by Monday | FSCS aims to return your money automatically within seven working days |
| Unauthorised transfers | Under Regulation E, your liability for an unauthorised electronic transfer is capped at $50 if you report it within two business days | Under UK payment rules, your bank must generally refund unauthorised payments promptly unless you acted fraudulently or with gross negligence |
Two points matter. First, the limit is per bank (or per banking licence in the UK): money spread across genuinely separate banks is protected separately, but money in two accounts at the same bank is not. Second, a transfer by an agent you authorised may not count as "unauthorised" at all — the gap Wells Fargo's new agreement points to, and the single most important thing to understand here.
Seven steps to keep your money under your control
You do not need to stop using AI tools or pull your cash out of the bank. You do need to decide, deliberately, what any software is allowed to do with your money.
1. Never give an AI agent your banking password
If an agent asks for your login, it is asking for the keys to your money with no limit on what it can do once inside. Use read-only connections where they exist — UK open banking lets apps see your balance without ever holding your password — and refuse anything that wants full login credentials.
2. Audit what is already connected to your accounts
Check the connected-apps or authorised-devices section of your bank's app and revoke anything you do not recognise or no longer use — and do the same for budgeting apps, trading apps and browser extensions. Access you granted an AI assistant "just to try it" may still be live.
3. Switch on passkeys or biometric login
A password can be handed to an agent — or stolen by one — while a fingerprint or face scan cannot be pasted into a prompt, which is why JPMorgan Chase now recommends passkeys or biometric sign-in. Turn both on wherever your bank offers them.
4. Turn on transaction alerts for everything
Enable push notifications for every login, password change, new payee, and any transfer at all. An agent moving money silently is the danger; an agent moving money while your phone buzzes is a problem you catch in seconds.
5. Keep your spending money separate from your savings
Hold day-to-day money in one account and the bulk of your savings in another, ideally at a different bank, with no debit card and no third-party access attached. Even in the worst case, the damage is contained — and this keeps you inside the deposit-protection limit at each bank.
6. Read your bank's updated terms before 2026 ends
Banks are rewriting their access agreements right now — Wells Fargo's takes effect on 24 November 2026, and others are following. Search your bank's terms for "agent" and check who bears the loss when an authorised tool misbehaves. If the answer is "you do", tighten the access in step 2.
7. Know exactly what to do if a transfer looks wrong
Call your bank immediately, using the number on the back of your card — not one from a text or email. In the US you can also complain to the Consumer Financial Protection Bureau; in the UK, to the Financial Ombudsman Service. The faster you report, the stronger your protection: under Regulation E, reporting within two business days caps your liability at $50.
Don't "fix" this by hoarding cash at home. The American Bankers Association points out that stashing physical cash trades one set of risks for another: money under a mattress can be stolen or destroyed by fire or flood, and it carries no insurance. Money in an FDIC-insured bank is covered up to $250,000 (or £85,000 under the FSCS in the UK) if the bank fails. A small emergency cash stash is sensible; your life savings in a drawer is not.
Barr's proposed guardrails — strict authentication, explicit consent for each specific action, technical limits on what an agent may do — point to what comes next. Until they exist in every bank's terms and every agent's design, the guardrail is you: decide what software may touch your money, check what already can, and make sure your phone tells you the moment anything moves.
Frequently asked questions
Has an AI agent actually stolen anyone's money?
Not that any regulator or bank has reported. What banks have reported is agents attempting actions the bank could not verify as authorised, and agents acting beyond what customers asked. The theft scenario is a warning, not an event.
Has AI caused a bank run?
No. Barr warned that agents could accelerate outflows during a stress period so a run unfolds in minutes. The last real run — Silicon Valley Bank in March 2023, $40 billion out in one day — was driven by humans. AI would make the same dynamic faster, not different in kind.
Is my money still insured if an AI agent moves it?
Deposit insurance covers you if your bank fails — up to $250,000 in the US (FDIC) or £85,000 per banking licence in the UK (FSCS). It does not cover money that leaves your account because you authorised a tool to move it: if an agent you authorised makes a transfer, your bank may treat it as your own action.
Key takeaways
- Fed Governor Michael Barr warned on 16 November 2026 that major banks are already finding AI agents attempting actions in customer accounts that the bank cannot verify as authorised.
- His core fear is speed: agents could detect trouble and move deposits so fast that a bank run unfolds in minutes.
- At least five large US banks — JPMorgan Chase, Bank of America, Citi, Morgan Stanley and Wells Fargo — are already blocking or restricting outside AI agents.
- Wells Fargo's updated agreement, effective 24 November 2026, treats an authorised agent's actions as your own.
- Deposit insurance still protects you if a bank fails ($250,000 FDIC; £85,000 FSCS per banking licence), but not transfers you authorised an agent to make.
- Protect yourself: never give an agent your banking password, audit connected apps, switch on passkeys and alerts, and keep savings in a separate account with no third-party access.
The Fed's warning is not a reason to panic — it is a reason to take thirty minutes this weekend and check what can touch your money. Agents are already inside the banking system; the question is whether yours answers to you, and only you. Open your bank's app, review the connected apps and authorised devices, switch on passkeys and alerts, and make sure no tool you tried once still holds the keys. Do it before the next agent asks for your password.
Sources
- Federal Reserve — Speech by Governor Barr on AI and financial stability
- Moneywise — Top Fed official warns AI bots could pull money out of banks
- CoinDesk — Autonomous AI agents risk triggering rapid bank runs, Fed Governor Barr warns
- TFTC — Fed's Barr: AI Agents Could Trigger Bank Runs in Minutes
- FSCS — If your bank fails
- AINewsIQ — Fed's Barr Warns AI Agents Could Speed Bank Runs
- The Life DAO — Can AI Agents Cause a Bank Run?
The MoneyHabit editorial desk researches, writes and reviews every article against our editorial policy.
The Sunday Habit
One useful money idea, every week.
A short, practical email every Sunday — no spam, no selling your address, unsubscribe anytime.
Advertisement
More in Financial Technology

Financial Technology · Guide
How to Close Old Bank Accounts Safely
Old bank accounts quietly cost you fees, leak your data, and can end up with the state. Here is how to close them safely — in the right order.
20 November 2026 · 12 min read

Financial Technology · News
Binance Is Closing EU Accounts on 31 Dec: Move Your Money
Binance will close accounts in France, Spain, Italy and Poland on 31 December — and charge a fee on leftover balances from 29 November. What to do now.
12 November 2026 · 8 min read

Financial Technology · Explainer
What Happens to Your Money When Your Bank App Goes Down
When your bank app goes down, your money is still there — but you cannot reach it. Here is what an outage breaks and how to build a backup plan.
7 November 2026 · 11 min read