Bank of America pays 40 cents to keep $1,000. AI may finally make customers move it

Bank of America will pay you 40 cents a year to keep $1,000 in its standard Advantage Savings account.

That’s not a typo.

The current rate is 0.04%.

For decades, banks could get away with rates like this because moving money was annoying.

You had to find another bank.

Open an account.

Transfer the money.

Remember passwords.

Move it back when you needed it.

For a few hundred dollars of extra interest, most people simply couldn’t be bothered.

That laziness has value.

Banks have built enormous businesses on it.

Apollo Global Management’s chief economist Torsten Sløk just pointed out what happens when software removes the laziness.

He calls it an “agentic bank run.”

The idea is simple.

An AI agent continuously looks at your cash and asks whether it could earn more somewhere else. Apollo says households could automatically move money from accounts paying around 0.1% into alternatives offering 3.3% to 5%. If enough households do it simultaneously, banks lose the cheap deposits they use to fund lending.

The strange part is that nobody has to panic.

Nobody has to hear that a bank is failing.

Nobody has to stand outside a branch.

Nobody even has to make a decision.

The machine can simply find a better rate.

That changes the meaning of a bank run.

The classic bank run is driven by fear.

The new version could be driven by optimization.

And someone actually described this before Apollo did.

A paper published in Finance and Society last year used the title “When Siri becomes a deposit broker.” It argued that AI agents instructed to maximize yield could move deposits from one bank to another even when nothing had gone wrong at the original bank. The paper also warned that agents built on a small number of common AI models could react similarly to the same information, creating herding.

That’s the part I would pay attention to.

The banking system doesn’t need millions of customers to suddenly become sophisticated investors.

It only needs millions of customers to give the same instruction to machines:

Get me the highest safe return on my cash.

The human doesn’t have to understand Treasury bills.

The human doesn’t have to compare APYs.

The human doesn’t even have to know which bank has the best rate.

The agent does it.

And suddenly Bank of America’s 0.04% becomes a problem.

Not because 0.04% is newly bad.

It has been bad for years.

The difference is that the cost of escaping it is approaching zero.

That’s a direct threat to one of the oldest advantages in retail banking: customer inertia.

Look at the scale of the system.

FDIC-insured commercial banks and savings institutions held about $19 trillion of deposits in Q2 2026, including about $4.0 trillion of noninterest-bearing deposits.

That deposit base isn’t just a pile of customer cash.

It’s bank funding.

And banks don’t all pay the same price for it.

A customer leaving $20,000 in a checking or low-rate savings account may be almost free funding for the bank compared with money that has to be attracted with a higher rate.

That spread is part of the business.

Now imagine an agent sitting between every household and every bank.

The agent doesn’t care about branch locations.

It doesn’t care that your family has banked there for 20 years.

It doesn’t care that your debit card is already in your wallet.

It sees 0.04%.

Then it sees 4%.

Then it moves the money.

The bank’s relationship with the customer suddenly becomes a price comparison engine.

And banking has already shown us how quickly deposit flight can happen once customers coordinate.

The Federal Reserve Bank of New York’s research on March 2023 identified 22 bank runs, not merely the two banks that ultimately failed. It found that relatively few but large depositors moved money quickly, with public signals helping coordinate the withdrawals.

AI could take that coordination mechanism and automate it.

That’s the uncomfortable part.

You don’t need everyone to independently decide that Bank A is unsafe.

You can have millions of agents independently following the same optimization rule.

The result can look coordinated even though nobody coordinated anything.

And it gets stranger.

The AI doesn’t need a banking crisis to move money.

A bank could be perfectly healthy.

Its competitor could simply offer 4.5% instead of 0.04%.

If the agent is told to maximize yield, that alone is enough.

That’s why this isn’t really a story about AI causing bank runs.

It’s a story about AI turning deposits from sticky relationships into constantly repriced inventory.

Banks have spent decades building branches, apps, credit-card relationships and loyalty programs around the assumption that customers won’t constantly shop their cash.

An AI agent can shop it every minute.

The irony is sitting right there in Bank of America’s rate.

The bank doesn’t need to convince you that 40 cents is a great return.

It only needs you to be too busy to move the money.

For years, that was a safe assumption.

AI may be about to make it obsolete.

Not financial advice.