BankingJPMOCT 18, 2026

Wall Street blocks third-party AI agents from moving money in four-day earnings sweep

JPMorgan and Schwab barred outside agents from logging into customer accounts on back-to-back Q3 calls, pointing agent-based payment flows at a read-only ceiling.

BY SEBASTIAN PHAM · BANKING & FINANCE AI

Over four days in October, the two most-watched Q3 earnings calls on the retail-finance tape converged on the same policy: outside AI agents won’t be allowed to log into customer accounts and move money. JPMorgan said it on October 13. Charles Schwab said it on October 14. By the time Rick Wurster sat down with Bloomberg Television on October 16, the posture had hardened into a weekly operations number.

Jamie Dimon, framing the call as a question of exposure, said JPMorgan won’t let “just any … outside agent to come in inside the company, because it’s just too risky for clients and for JPMorgan.” A JPMorgan spokeswoman described what the bank is specifically blocking third parties from doing: “obtaining customers’ banking credentials and logging in on their behalf.” No timeline for an approved channel was offered. Agents can still see balances through authorized connectors like Plaid. They can’t send money or pay bills.

Schwab’s version arrived a day later and came with a count. Wurster told the earnings call “we are blocking unauthorized agents who try to log in on behalf of clients,” then told Bloomberg that outside providers, including Meta’s Muse, are stopped “in the tens and twenties” each week. His stated concern, per Yahoo Finance, was accountability: “something could go wrong and who’s liable?” Dimon had already supplied the phrase, “liability transfer.”

The replacement architecture is already visible. Schwab plans to expose read-only account data and research to outside LLMs through bank-run MCP servers, with a trading pilot limited to select clients and models. Money movement stays inside Charley, Schwab’s own assistant. OpenAI’s Dots, Meta’s Muse, and Spear Street’s Instinct end at a read ceiling.

For a 5-to-30-person company that budgeted on autonomous payment agents clearing invoices overnight, the design implication is immediate. Credential-sharing automations need auditing. Agent permissions split cleanly into read, trade, and money-movement tiers. A named human releases payments. The echo of our earlier read on BofA’s railroad analogy is unmistakable: the incumbents are drawing the gauge of the track, and nothing either CEO said exempts commercial portals.

Sources

About the author

Sebastian Pham

BANKING & FINANCE AI

Sebastian Pham reports from the London bureau on how the largest banks, insurers, and capital-markets firms are deploying agentic systems. His beat is bank-specific: headcount shifts, ops workflows, vendor selections inside tier-one institutions.