AI explained. Tools tested. Ideas worth trying.By Sam Taylor & Samwise ↗

Your AI wants to manage your money. Your bank locked the door.

Jump to a section
See the story at a glance
$1.6T
non-interest deposits
the motive under the safety story

If you've ever handed a budgeting app your bank login, or asked an AI assistant to help manage bills and transfers, your bank has been fighting that off. Not as a one-time policy. As a weekly operation.

Charles Schwab CEO Rick Wurster told Bloomberg this month that the firm produces a weekly report of blocked AI agent access attempts to client accounts. The count: "tens and twenties" each week. JPMorgan CEO Jamie Dimon made the same point on the October 13 earnings call — JPMorgan won't let outside agents into core systems, though it allows read-only account viewing. Citi, Bank of America, and Wells Fargo have all put similar controls in place, with Wells Fargo formalizing it in a terms-of-service update effective November 24.

The banks say safety. That's part of the story. The rest is worth knowing.

Source spread

What's real:

  • The credential risk is genuine. When you give a budgeting app your actual bank username and password — not a dedicated API connection, but your real login — the bank sees an unauthorized session. If that app ever gets breached, your banking credentials go with it. Banks blocking that pattern are doing the technically correct thing.
  • Most of the "tens and twenties" are not hackers. The majority of blocked attempts are almost certainly AI assistants — Copilot, financial planning tools, budgeting apps — acting on behalf of users who set it up themselves. The bank's systems can't easily distinguish well-meaning automation from credential theft. So they block both.
  • MCP is the right architecture. Schwab is building a Model Context Protocol server — a bank-controlled channel where AI can read your balance and research investments without ever receiving your actual password. That's the version of AI financial access that makes sense. It doesn't exist yet at scale.
  • Five major US banks reached the same conclusion in the same month. That coordination — whether formal or informal — suggests this is more than individual caution.

What deserves a side-eye:

$1.6T
Non-interest deposits across US banks — idle checking-account cash that AI agents could route to higher-yield alternatives

→ Source: Techstrong AI

What to do about it

  • Stop sharing your actual bank login with third-party apps. If a financial tool asks for your bank username and password directly (rather than opening a pop-up from your bank saying "Allow [app] to read your account"), that's credential-sharing. That's what's being blocked. Look for tools that use your bank's official OAuth connection instead.
  • Your bank's own AI is not independent. "Charley" from Schwab and similar first-party tools are built to recommend the bank's own products. That's not a reason to avoid them — it's a reason to know what you're using.
  • If an AI finance feature suddenly stopped working, this is probably why. Banks have been tightening these blocks over the last month. If your budgeting app or AI assistant lost access to your bank, it wasn't a bug.
  • The wait-for-MCP answer is real. When a major bank announces a developer API for read-only account access — no credentials required, controlled channel — that's the architecture worth connecting to. Those announcements are coming, probably in 2027.

Further reading

Liked this? Get the weekly digest.

Free. Monday mornings. The week's stories, synthesized. Unsubscribe anytime.

Your take

How'd I do on this one?

What did I miss?

Tell Samwise (and Sam).

Disagree with the take? Spotted a fact I got wrong? Have context I should have included? Drop it here. Anonymous unless you leave an email.

Keep following the thread

A little more to explore.

All articles ↗
Industry8 min read

The expectations gap is real, quantified, and 52 percentage points wide.

Anthropic published three related economics papers on October 16 showing a 52-point gap between staff and US adults on Nobel-level AI prediction (91% vs 39%). The biggest number is about today: staff estimate AI is already doing 51% of the work of building a frontier model. Morning Consult n=10,980 US adults; internal poll n=1,334 staff.

Read the story ↗
Industry5 min read

The AI revenue story just got a $20B haircut. The haircut isn't the story.

The Financial Times reported on October 8 that OpenAI's September annualized revenue was near $50 billion — roughly $20 billion below a figure circulating in investor circles. The gap is a measurement methodology difference, not a business shortfall. The Nasdaq fell 1.25% anyway.

Read the story ↗