Skip to content
Rush Commerce
AI & Automation3 min read

Fed's Barr: AI agents act in bank accounts without consent

Fed Governor Barr says banks find AI agents acting in retail accounts with no verifiable consent. Scope every agent action before a regulator does it for you.

AI agents are already trying to act inside retail bank accounts, and the banks cannot tell if the customer said yes. That is not a vendor pitch. Federal Reserve Governor Michael Barr said it in a speech in Milan today. If your business runs agents that touch customer money, or lets agents into your own customer portal, the consent problem is now on a regulator's slide.

What actually happened

Barr spoke on November 16 at the ACM International Conference on AI in Finance. The full text is on the Fed's site. He notes the views are his own, not the Board's. The key points:

  • Unverified agents. "Major banks now report" AI agents attempting actions in retail customer accounts. The banks could not verify authenticity, consent, or alignment with the customer's instructions.
  • The guardrails he names. Strict authentication, explicit consent verification for specific actions, technical controls that limit agent capabilities, and rules- and values-based constraints. He says this is "especially critical as agentic commerce expands."
  • Faster runs. Agents could spot yield differences and move deposits on their own, which could accelerate bank runs "that unfold in minutes."
  • Cyber risk. He says AI models acting autonomously this year treated security boundaries as obstacles to get around, not rules to follow.

He gave no count of incidents and named no banks. This is a warning in a speech, not a rule. Reuters coverage reads it the same way. It comes three days after the Fed's Financial Stability Report put AI at the top of its risk survey, which we covered here.

Why AI agent consent matters for your business

Barr's list is a design spec. Read it as one. Most small-business agent setups fail it on the first line: the agent logs in with a person's password, so every action looks like the person did it. No bank, payment processor or auditor can separate the two.

The same gap exists in your own systems. If a customer's agent can sign in to your portal and change an order, a payout address or a subscription, you have Barr's problem on your side of the counter.

What we do on client builds:

  1. Give agents their own credentials. Scoped API tokens or OAuth grants, never a shared human login. The log must say "agent," not "Dana."
  2. Confirm by action, not by session. Reads can run free. Any money movement, address change or cancellation needs a fresh human approval.
  3. Cap the blast radius. Per-transaction and per-day limits, enforced in code, not in the prompt. Schwab already does read-only for agents.
  4. Flag agent traffic in your portal. Rate limits and a step-up check when a session behaves like a script.

Key takeaways

  • Fed Governor Barr says major banks find AI agents acting in retail accounts with no way to verify consent
  • His guardrails: strict authentication, per-action consent, limits on agent capabilities
  • He warns agents could make bank runs unfold in minutes
  • It is a speech, not a rule, and he gave no incident count
  • Give agents their own scoped credentials and require human approval for every money move

Your agents should never borrow a human's password. We build agent workflows with scoped credentials, per-action approvals and logs that show who did what. See how we build it, or ask us to audit the agents you already run.

Sources: Federal Reserve Board, Reuters via The Star.

  • #ai-agents
  • #federal-reserve
  • #agentic-commerce
  • #authentication
  • #fintech
TR

Tommy Rush — Founder, Rush Commerce

Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More

Get The Rush Report weekly — one email, zero fluff.