Big banks are shutting the door on AI agents: the problem is identity, not AI
Direct answer
According to Reuters on October 22, 2026, major U.S. banks are blocking third-party AI agents from operating inside customer accounts and making transactions. The report names Morgan Stanley, JPMorgan, Bank of America and Citi, with Wells Fargo taking its own steps per Axios, and Charles Schwab's chief executive confirming his firm turns agents away too. The banks' stated reason is threefold: credentials, permissions and financial data exposure. The nuance matters: JPMorgan and Citi allow some approved read-only connections, according to follow-up coverage. This is not a blanket ban on AI in banking: it is a ban on agents that cannot prove who is acting, under whose authority and with what scope.
Why banking is the first wall for agents
- Money is the task where a mistake costs most: an agent that misreads an email produces a bad draft. An agent that moves money with broad permissions produces a real loss and a real claim.
- The shared password is the original problem: many agents work with the user's credentials, as if they were the user. To the bank, that session is indistinguishable from impersonation.
- Regulators were already watching: the same month, the FTC sent investigative demands to OpenAI and Anthropic and the FSB published practices for financial institutions using AI. Banks are not acting in a vacuum.
- Agent identity does not yet exist as a standard: without a verifiable agent identity, scoped permission and automatic revocation when behavior drifts off-script, a bank has only two options: let everything through or block everything. Several chose to block.
What to do if your business wants an agent to touch money
- Start with read-only and approval: check balances and transactions through an approved connection, and keep payments and transfers behind explicit human approval.
- Demand identity and scope in writing: which agent, acting for whom, with which permission, over which accounts and with which limit. If your vendor cannot document it, do not connect it to real money.
- Prepare the revocation plan: a button and a procedure to cut the agent's access in minutes, not a ticket that takes days.
- Log every agent action: the log is the evidence the bank, the auditor and the client will ask for when something goes wrong.
The honest note
- The report rests on sources familiar with the situation: Reuters cited two people familiar with it. The banks did not publish a joint policy or a list of blocked agents.
- Positions differ by bank: blocking, own steps and approved read-only connections all appear in the same coverage. Presenting it as a uniform ban would be inaccurate.
- Sources: Reuters (October 22, 2026) via Tech Insider and Superpower Daily follow-up coverage (October 23, 2026).
Want AI that works for your business?
Message us on WhatsApp and let's talk about automation and AI agents for your business.
WhatsApp inicIARelated articles
- The FTC asked OpenAI and Anthropic for 50+ pages of documents
- 1Password for Agents: your agents sign in without ever seeing your password
- AEREDIUM launched controls for agents that move money
Frequently asked questions
Do all banks block all AI agents?
No: the report describes blocks on third-party agents in accounts, with nuances. JPMorgan and Citi allow some approved read-only connections, according to coverage.
Why do banks block agents?
Over credentials, permissions and financial data exposure: they cannot verify which agent acts, under whose authority and with what scope.
Can I use an agent to check my account?
It depends on the bank and the connection. Approved read-only connections are the starting point several banks do accept.
What does an agent need to operate with money?
Verifiable identity, scoped permissions, human approval for payments, and a fast revocation mechanism.
