AI agents are getting surprisingly good at handling tasks nobody particularly enjoys. They can compare prices, organize schedules, research investments, and analyze spending habits.
Ask one to move $5,000 into a better savings account, though, and things get considerably more complicated.
JPMorgan Chase and Charles Schwab are restricting what outside AI agents can do with customers’ financial accounts. Both companies support AI-assisted financial services, but neither wants unauthorized third-party software independently controlling customer funds.
Chase allows certain financial information to be shared through approved connections, while Schwab is developing controlled ways for customers to access account data using AI.
Actual transactions are another matter.
For AI companies hoping to build assistants that handle practically everything, banking is proving to be a particularly stubborn obstacle.
In this article
What Chase and Schwab Are Actually Allowing
JPMorgan CEO Jamie Dimon addressed the issue during the bank’s October 13, 2026, earnings call, warning about the security risks associated with outside AI agents.
His concern involves potentially thousands of unauthorized or malicious agents attempting to access financial accounts.
Chase already blocks external AI agents that try to sign in using customers’ banking credentials and operate accounts on their behalf.
However, authorized connections through financial data providers such as Plaid can access certain account information, including balances, on a read-only basis.
That distinction matters.
An AI assistant connected through an approved service might help analyze someone’s spending or identify idle cash. It cannot independently initiate payments or transfer funds using that connection.
Charles Schwab is taking a similar approach, although its specific rules differ.
CEO Rick Wurster has expressed skepticism about letting AI independently make major investment decisions. Schwab is also blocking unauthorized agents attempting to access customer accounts.
At the same time, Schwab is developing connections through Model Context Protocol (MCP), a system that allows AI applications to communicate with external tools and data sources under defined permissions.
Your AI can help figure out where your money should go. Actually sending it there is where the humans, or at least additional authorization controls, get involved.
Why Banks Don’t Want AI Agents Touching the Money
Security is the obvious concern, and it is not particularly difficult to imagine what could go wrong.
AI agents can misunderstand instructions, encounter manipulated information, or execute actions their users never intended.
Researchers have already demonstrated how troublesome that last possibility can be.
A 2026 security competition involving more than 250,000 attack attempts against 13 advanced AI models found at least one successful hijacking attack against every model tested.
Nobody was conducting actual bank transfers in these tests, but the underlying problem is uncomfortable enough: carefully crafted instructions can persuade an assistant to do something its owner never requested.
A chatbot incorrectly calculating an investment return is annoying. An autonomous assistant transferring $10,000 to an incorrect account is a much more expensive problem.
Accountability introduces another complication.
If an AI agent makes an unauthorized payment, responsibility could involve the customer, the AI provider, and the financial institution. Existing protections may not neatly address every scenario involving autonomous software.
However, security isn’t the only consideration.
Banks also have financial incentives to limit autonomous agents.
An assistant capable of continuously comparing savings rates could identify accounts paying very little interest and automatically transfer customers’ deposits to better alternatives.
For consumers, that sounds rather useful.
For banks relying on inexpensive deposits to fund lending activities, widespread automated transfers could become a serious business problem.
Of course, financial incentives don’t invalidate legitimate security concerns. Both considerations can exist simultaneously, even if one makes for a less flattering earnings-call discussion.
Banks Still Want AI, Just on Their Terms
Neither company is rejecting artificial intelligence.
JPMorgan is piloting Smart Cash, an internally developed tool designed to forecast cash flow and help optimize where customers hold their money.
Schwab has introduced Charley, its own AI assistant, which currently focuses on answering questions, synthesizing information, and handling selected administrative tasks.
Schwab is also experimenting with AI integrations for financial advisors and trading customers.
What’s different is the level of control.
Banks can establish authentication requirements, transaction permissions, and security restrictions within systems they manage.
An independent AI agent operating through someone’s login credentials creates considerably more uncertainty.
For now, neither institution has announced unrestricted transaction access for outside AI agents.
What Does This Mean for Customers?
Anyone hoping to connect an AI assistant to their financial life should distinguish between accessing information and authorizing transactions.
Approved integrations may allow assistants to review balances, analyze spending, identify savings opportunities, and support investment research.
Customers should not expect an independent agent to freely pay bills, move deposits, or execute trades through Chase or Schwab.
Future systems could introduce explicit approvals, transaction limits, and stronger identity verification before allowing sensitive actions. Neither company has committed to making unrestricted autonomous banking available.
And that leaves AI agents in an interesting position.
They can increasingly help customers decide what to do with their money, but actually doing it still requires permission.
Given how confidently chatbots can occasionally get things wrong, that might not be the worst arrangement.
