Audit your agent's logins in 20 minutes, before your bank does it for you

When you set up an Always On Agent to reconcile payouts or check balances, the quick way in was to hand it your bank login. That's the access banks have started shutting off.

If your agent signs in as you, the bank sees you. And the mistakes are yours.

What banks are doing

On Oct 22, Axios reported that Morgan Stanley has begun blocking third-party AI agents, and that JPMorgan, Bank of America and Citi are restricting them. Per Axios, most of what banks find is agents using a customer's own login with the customer's permission. That still breaks bank rules that say a human logs in. Banks often can't tell an agent the customer sent from a rogue one, so they block first and rewrite their terms.

Reuters followed the same day. Two people told Reuters that major banks have been blocking third-party agents from operating inside accounts and making transactions. When customers share credentials, banks typically flag and block those transactions. Schwab CEO Richard Wurster said earlier this month: "We are blocking it because we want to have control over authentication and the experience."

Axios also reported, from unnamed sources, that some agents were acting independently of user instructions. Treat that part as a claim. It's disputed. Critics say the story mixes up agents the customer authorized with rogue ones, and Morgan Stanley says it hasn't seen rogue agents. The blocking is the confirmed part, and it's the part that hits you.

Chase already wrote the rules

Chase's Terms of Use now cover AI agents. Any agent touching Chase systems has to identify itself as an agent at all times, "clearly and deterministically label all agentic traffic," answer truthfully if asked whether it's human, and stop when Chase asks. Chase can limit or end agent access at its discretion.

Then the line every owner should read twice. If the agent misreads your instructions, does something you didn't intend or goes past the authority you gave it, Chase says "you are responsible for any actions or errors that may result from the Agent's incorrect interpretation or implementation of your instructions."

An agent signing in to Chase as you, with no label, is outside those terms today. And the risk sits with you.

The audit, in six steps

Set a 20 minute timer. Open the agent's settings next to your password list.

1. List every login the agent holds

Bank, card portal, payroll, payment processor, marketplaces, ad accounts. Next to each, write how the agent got in: a password you pasted, a login saved in the agent app, an approved connection (OAuth) or an API key.

2. Sort each login into three tiers

• Read only. Balances, statements, reports.
• Act with approval. Pay a bill, issue a refund.
• Never. Move money between accounts, change payees, change login details.

Most of a bookkeeping agent's job lands in the first tier. Keep the third tier short and absolute.

3. Move bank access to read only

Axios reports most banks are allowing read-only access or moving toward it, the way budgeting apps connect. Use your bank's own approved connection where one exists. Stop giving an agent your bank password.

4. Rotate anything you pasted

If a password ever went into an agent chat or a settings page, change it and turn on two-factor. The old copy is now dead wherever it sits.

5. Read your bank's agent terms

Chase spells its rules out. Look for the same section at every bank and processor the agent touches. If a bank says humans only, follow it or move the task somewhere that allows agents.

Approved sign-in tools now exist, where the agent asks for a login, you approve on your own device, and the tool fills the form so the agent never sees the password. That keeps the password safe. It's still the agent working inside your session, though, and on our read it doesn't label the traffic as an agent on its own. Check that against terms like Chase's before you rely on it.

6. Gate the money moves

For anything in the "act with approval" tier, set the approval inside the system that moves the money. We mapped that for Shopify and Stripe in draw your agent's money map before you connect it. Give each agent its own login wherever the system allows one, the way we laid out in give each Always On Agent its own seat. And decide now what you'll do the day an approved action goes wrong. We wrote that drill in your agent pressed submit.

What this doesn't tell you

The Axios story leans on unnamed sources, and bank positions are moving. We're not naming which bank blocks what beyond what Axios and Reuters reported. Chase's terms page shows no date, so check it again before you act on it.

What it does give you is a list of every login your agent holds, how it got each one, and which tier each belongs in.

This week

Run the list. Move bank logins to read only. Rotate every password that ever touched an agent. Read the agent section of your bank's terms. Put approval on every money move that stays.

If you want your agents' access set up this way from day one, Start your build.

Linked sources

• Axios (Madison Mills): banks restricting third-party AI agents in customer accounts (2026-10-22)
• Reuters via US News: US banks detect unauthorized AI agents in customer accounts (2026-10-22)
• JPMorgan Chase: Terms of Use, AI Agent or Agentic AI Notices
• Related article: Draw your agent's money map before you connect it
• Related article: Give each Always On Agent its own seat, not your login
• Related article: Your agent pressed submit