Developments3 min read
Fed warns on AI agents: banks can't tell who is clicking

AI agents could accelerate bank runs that unfold in minutes, Fed Governor Michael S. Barr warned in a speech in Milan on November 16, 2026. Major banks report agents trying to act in customer accounts without any way to verify authenticity or consent, per Barr. For stores and brands there is a second warning in there.
Key facts
- Fed Governor Michael S. Barr warned in Milan on November 16, 2026, that AI agents could accelerate bank runs that unfold in minutes; he said the views are his own.
- Major banks report AI agents trying to act in retail customer accounts without a way to verify authenticity, consent or the customer's instructions, per Barr.
- Barr calls for strict authentication, explicit consent for specific actions and technical limits on agents, and warns of herding among similar AI models.
Researched and written with AI, to Luigi's brief, positions and editorial rules. Automatically checked against the sources. How I work →
Picture it. An order lands in your shop backend at 3:12 a.m., cleanly filled in and paid. No human clicked. An agent did. Did the customer actually ask it to buy exactly that? You can't tell. Today the US central bank said the big banks can't tell either.
What Fed Governor Barr said about AI agents
Michael S. Barr, a member of the Federal Reserve Board, spoke about AI and financial stability at an Association for Computing Machinery conference in Milan on November 16, 2026. He stressed the views are his own, not the Board's. The parts that matter for marketing and e-commerce:
- Major banks report AI agents trying to take actions in retail customer accounts, per Barr. The banks could not verify authenticity, consent, or whether the action matched the customer's instructions.
- Agents could spot yield differences or stability signals instantly and move money at high speed, so bank runs could unfold in minutes.
- Similar models trained on similar data can act in sync (herding). One shared flaw becomes a system failure.
- What he asks of firms: strict authentication, explicit consent for specific actions, and technical limits on what an agent can do.
The idea that agents could drain bank deposits has been around the finance press for a while. The two points I find more interesting sound like banking and are really about marketing.
Consent per action becomes a checkout problem
If a bank with a compliance department can't tell whether an agent acts on someone's behalf, your checkout certainly can't. To me that is the real bottleneck of agentic commerce: not the model, but proof that this customer wanted exactly this. The Fed line, consent for specific actions, will become the yardstick for payment providers and fraud filters. Translation: agents will meet friction. Stores that clearly describe ordering and returns for agents get preferred, the rest land in the review pile. It fits what I laid out in my timeline on ChatGPT Wallet and AI agents that pay.
Herding: when every agent thinks alike, they all switch at once
Barr means herding in trading. I'd apply it to brands. If your customers' shopping agents run on a handful of big models, they decide alike. If one model's answer about your brand flips, you don't lose one customer, you lose many at once. That's the opposite of the funnel we know, where customers drift away one by one. Agents are gatekeepers, and they move as a pack.
The catch
This is a speech, not a rule. Barr speaks for himself, not the Fed, and none of it applies directly in the EU. But the banks he cites are the same ones being asked to approve payments by agents. What they learn there ends up in your checkout. How fast agents turn into a liability case is also clear from California's lawsuit against OpenAI over AI agents.
What I'd do now
- Have someone check whether agent orders are recognizable in your shop backend and how your payment provider treats them. The AI Agent Radar shows where the big AI agents get in.
- Describe ordering, cancellation and returns so an agent finds them without guessing. The guide Marketing to AI agents shows how.
- Measure the herd risk: ask ChatGPT, Claude and Gemini the same customer question and check whether they all name the same competitors. The AI Duel does that automatically.
Bottom lineThe Fed fears agents pulling money in lockstep, and for brands it is the same risk with customers instead of deposits.
Got AI news? Submit news →
FAQ
What did the Fed say about AI agents and bank runs?
Fed Governor Michael S. Barr said on November 16, 2026, that AI agents could spot yield differences instantly and move money fast, so bank runs could unfold in minutes. He spoke for himself, not the Fed.
What does Barr ask of firms using AI agents?
Strict authentication, explicit consent for specific actions, and technical limits on what an agent is allowed to do.
What does herding mean for AI models?
Similar models trained on similar data can act the same way at the same time. Per Barr, one shared flaw can then become a system-wide failure.
Source
More sources
- Gizmodo – Hinweisgeber, Zusammenfassung nach Bloomberg
- Federal Reserve: Reden 2026 – Datum, Titel und Ort der Rede



