US banks block AI agents from accounts as regulators lag on liability rules
Five of America's largest banks — JPMorgan Chase, Citi, Bank of America, Wells Fargo and Morgan Stanley — are blocking external AI agents from reaching customer accounts, as the industry decides for itself how to police bots that spend on a consumer's behalf.
The stakes are two-sided. Banks fear fraud, because a request arriving from a customer's authorised agent can be near-impossible to tell apart from one sent by a bad actor, said Justin Herring, a partner at law firm Mayer Brown. They also worry about faster deposit outflows once a bot can move money at speed. Herring said that from a technical standpoint it can be "impossible" to determine whether a request comes from a threat actor or an actual customer.
Liability is the harder knot. Existing rules generally shield consumers from unauthorised transactions, said Julie Hill, dean of the University of Wyoming College of Law — but an agent acting on a customer's instruction muddies what "authorised" means. If a bank cannot show a transaction was authorised, its exposure rises. JPMorganChase has said customers who hand authority to an AI agent must understand the liability that comes with delegating it.
With no rulebook yet, banks are expected to build their own guardrails: asking customers to disclose agent use and confirm specific transactions, and working out how an agent proves it was authorised. David Zaring of Wharton said the agents are already here while rulemaking takes time, leaving banks "walking the tightrope without much of a net".
Prudential regulators are watching. Comptroller of the Currency Jonathan Gould said in late October that banks are "some of the most fortified parts of our system" and are focused on defending the frontier. Herring expects rules of the road eventually, noting they usually lag industry practice — and, as Hill put it, banks, customers and merchants all have reasons not to wait.
What it means for Pakistan: Pakistan's banks are only beginning to wire AI into customer-facing services, and none of the large US institutions' block-on-agents stance binds them — but the liability question lands here too. The State Bank of Pakistan's consumer-protection and electronic-payments rules were written for transactions a human initiates, and the country's fraud-reimbursement practice is thinner than the US framework the article describes. As Pakistani banks and fintechs experiment with chatbots and agent-style payments, they will face the same problem of proving who authorised a transaction, with less case law and no dedicated AI-agent guidance from the State Bank or SECP to fall back on.
