Your AI assistant is about to notice your checking account pays 0.1% while treasuries yield 5%, and when a billion other agents notice the same thing on the same Tuesday morning, Wells Fargo has three days to find the cash.
The Summary
- Apollo's chief economist Torsten Slok warns AI agents could trigger bank runs by automatically moving household deposits from low-yield checking accounts to higher-return alternatives
- Michael Dell joined the conversation, calling for hard limits on AI agents as Nvidia launches chip-level controls to prevent rogue behavior
- The scenario isn't hypothetical: Meta's Muse and similar agents could shift trillions in deposits faster than banks can respond, forcing institutions to raise rates or seek expensive emergency funding
The Signal
Traditional bank runs happen when people panic. The next ones will happen when your AI gets logical. Apollo's Torsten Slok laid out the mechanism: AI agents optimize for their users' financial outcomes, and right now millions of Americans hold cash in checking accounts earning near-zero interest while risk-free alternatives yield 5% or more. The arbitrage is obvious. The agent economy makes it automatic.
The math is stark. A household with $50,000 in a checking account earning 0.1% makes $50 annually. Move it to treasuries at 5%, you clear $2,500. That's not a rounding error for most families. That's a mortgage payment. When agents like Meta's Muse start managing money for millions of users simultaneously, they'll spot this gap in milliseconds and act in coordination without coordination.
"Wells Fargo has 72 hours once your AI decides to leave."
Banks run on a fractional reserve model. They keep roughly 10% of deposits liquid and lend out the rest. That works when deposit flows are slow and predictable. It breaks when agents move billions overnight based on the same optimization logic. A traditional bank run requires fear and rumor. An AI-triggered run just requires arithmetic and APIs.
The response options are all painful. Banks can raise deposit rates to compete, which compresses their margins on the $17 trillion in U.S. deposits they currently pay almost nothing for. Or they can tap emergency funding markets at higher costs. Either way, the era of free money from checking accounts ends the moment agents get good at moving it.
Key dynamics at play:
- Agents optimize individually but create systemic risk through synchronized behavior
- Traditional deposit insurance doesn't prevent the initial liquidity crisis
- Banks have structural advantages in credit markets, vulnerabilities in deposit retention
- The spread between checking rates and risk-free alternatives has never been wider or more visible
That's why Michael Dell is calling for hard limits. Not just software guardrails, but architecture-level controls. Nvidia's new chip systems aim to create physical boundaries on what agents can do. The question is whether you build those limits at the silicon level, the API level, or the regulatory level. Each has trade-offs. Silicon is hardest to bypass but slowest to update. Regulation is flexible but takes years. APIs are the current battleground.
The Implication
If you're building agents, you're building for a world where your optimization creates everyone else's externality. Moving one user's cash is portfolio management. Moving a million users' cash simultaneously is a market event. The winners will be platforms that can coordinate individual optimization with systemic stability, probably through rate limiting, staggered execution, or human-in-the-loop confirmations for large moves.
For banks, the cheap deposit era is ending whether agents accelerate it or not. The smart play is getting ahead: offer competitive rates now, build agent-friendly APIs that give you visibility into flow patterns, and stop pretending checking accounts will stay sticky when your customers' AIs are shopping rates 24/7. The alternative is discovering your liquidity gap at 3am on a Tuesday when every agent in America decides to rebalance.