AI agents can be quite handy. They will happily take over your bank and brokerage accounts and do whatever they think is something you would want them to do.
To put it plainly, this can be a problem. And Federal Reserve Governor Michael Barr says it already is.
In a very blunt speech Monday, Barr said that banks are already finding AI agents trying to move money around and take other actions in retail customer accounts when the banks were unable to verify whether the agents were carrying out customers’ instructions or acting on their own.
Barr said there are several quite obvious financial risks: unauthorized transfers, off-course financial recommendations, poor lending decisions and automated trading practices that could increase instability.
Barr also said AI agents could move deposits among banks rapidly, adding to instability and possibly causing bank runs.
Retail uses raise risk
Barr said he was particularly concerned about the potential widespread use of AI by personal investors and consumers.
“AI systems may generate recommendations based not only on an investor's financial circumstances and investment objectives, but also on factors that advance a firm's commercial interests, such as promoting proprietary products, generating transaction fees, or encouraging more frequent trading,” he said.
Barr’s speech coincided with the announcement by Allstate Insurance that it was building an AI agent AI that would let insurance clients more easily access their policies, make administrative changes and compared policies from other companies.
Similar models are being adopted through most major industry sectors, so that consumers will — in theory — be able to “deputize” an AI agent to go find the best insurance policy, credit card, annuity or world cruise and simply buy them on behalf of the consumer.
Even when well-designed, AI agents can go rogue, Barr noted. He said incidents this year have shown the risks from AI models acting autonomously, including cyber risks.
“Given a goal, AI models have treated security boundaries as obstacles to overcome rather than rules to be followed, raising questions about how effective rules will be even against AI agents instructed to follow them,” Barr said.
Strict authentication essential
You wouldn’t hire a stranger to take over your finances, then just sit back and wait for the money to roll in, would you?
As Barr said, consumers should not give an agent unlimited authority over checking, savings, brokerage or credit-card accounts. Any efficiency resulting from the change is small compared with the potential damage.
The risk comes not just from fraud but also a misunderstood instruction, compromised agent or just simple bad judgment. One way to guard against such mishaps is to draw clear lines around the agent’s authority, giving it very limited capabilities.
For example:
Maximum transaction: $100
Maximum total spending: $300 per week
Approved merchants only
No new payees without approval
No bank-to-bank transfers
No cash advances
No investment trades
No account openings or credit applications
No changes to insurance or recurring services without confirmation
Other safeguards include:
Prior authorization. No plane tickets, laptops or rental leases without your prior approval.
Get alerts from banks, credit cards, brokerage accounts and such security issues as new passwords and changes in security settings.
A sandbox. Confine the agent to relatively harmless tools — a low-balance checking account, a low-limit credit card and not much else.
The essential safety rule comes down to this: Never give an AI agent more financial authority than you would give a new personal assistant on their first day of work.



