Why Banks Are Blocking Outside AI Agents

What the reporting supports

October reporting describes JPMorgan Chase and Charles Schwab restricting unauthorized outside AI agents from accessing customer accounts or moving funds. The Next Web, citing Schwab CEO Rick Wurster’s Bloomberg Television interview, names Meta’s Muse among the agents being blocked. It also reports Jamie Dimon’s objections to outside agents entering JPMorgan’s systems. These are institution-specific restrictions, not evidence that every U.S. bank has adopted an identical policy. [1]

Reading information and moving money are different permissions

The same report describes Schwab’s proposed controlled access for balances and research, while withholding permission to move money. It also distinguishes JPMorgan’s approved data-sharing arrangements from unrestricted outside-agent access. The scope matters: an account connection can expose information without authorizing a transaction. [1]

This is also a story about who builds the assistant

Schwab’s own September 30 announcement introduces Charley, an assistant rolling out in stages beginning in October. Its stated functions include answering account questions, surfacing market information and completing selected account-maintenance actions. Schwab also says the assistant can produce inaccurate or incomplete results and asks users to verify information and completed changes. Its announcement supports a distinction between institution-managed AI and outside automation; it does not establish unrestricted autonomous trading. [2]

The question of responsibility

Wurster’s stated concern centers on control and responsibility when a customer gives an agent access and something goes wrong. [1] MKP’s interpretation is that a useful financial assistant needs a clear boundary between suggesting an action and having authority to carry it out. Customer intent, service permission and actual execution must line up.

Consider an illustrative request: “Find a better place for my savings.” Does that mean compare accounts, prepare a recommendation, open a new account or transfer the balance? Each step grants more power. This example is a design question, not an account of a documented incident.

Security and business incentives can coexist

MKP’s analysis: independent assistants could make comparing providers easier and reduce the friction of switching. That creates a possible commercial incentive to keep customers inside a provider’s own tools. But an incentive is not proof of motive, and the cited statements do not establish that protecting profits is the sole or primary reason for the restrictions. Customer protection and competitive interests deserve separate examination.

What would earn trust?

Our editorial criteria for future systems are concrete: understandable permissions, separate approval for consequential actions, a record of what was authorized and performed, a way to revoke access, and an explicit process for resolving mistakes. These are proposed evaluation criteria, not a claim that every current product provides them or that any technical label guarantees safety.

Why this belongs on MKP

AI’s progress is measured partly by what it can do. This story asks what people and institutions will allow it to do. The next stage of automation will depend on trustworthy delegation as much as model capability.

Discuss

• Should an assistant be able to compare accounts while requiring a fresh approval for each transfer?
• How should customers distinguish a security restriction from a commercial barrier?
• What evidence would make you comfortable delegating a consequential task?

Join the MKP discussion →

Sources and evidence limits

1. The Next Web, October 16, 2026: Schwab and JPMorgan are blocking AI agents from moving customer money. Secondary reporting on executive statements; the linked Bloomberg interview report was not accessible for direct review.
2. Charles Schwab, September 30, 2026: Introducing Charley. Primary source for Schwab’s stated product capabilities and limitations.

Original MKP synthesis. Reported facts, illustrative examples and editorial analysis are identified separately. Policies and product capabilities may change.