Torsten Slok, chief economist at Apollo Global Management, warned this week that autonomous AI agents could trigger a new form of financial instability: an "agentic bank run." Slok's analysis, published in a research report on Sept. 27, projects that software agents would automatically move household cash to maximize returns at machine speed and scale. This mechanism, while framed for traditional finance, already operates within decentralized finance.
Unlike traditional bank runs driven by a loss of confidence and panic withdrawals, an agentic run stems from optimization. Agents follow instructions to chase the best available yield continuously, without human hesitation or inertia. If millions of these agents act on the same signal, the resulting capital outflow would resemble a coordinated drain rather than a crowd reaction.
Traditional deposit funding relies on human friction, where depositors move money slowly, tolerate low interest, and rarely reallocate funds all at once. This stickiness allows banks to lend long against short-term deposits. Removing this friction by AI agents makes deposits highly mobile, transforming them into the most fluid money in the financial system.
The dynamics Slok describes are not a future hypothesis in crypto. On-chain, agents continuously move money to the highest yield instantly. Automated vaults, yield routers, and rebalancing bots shift stablecoin liquidity between protocols the moment rates diverge. This constant optimization is a core function of a large share of decentralized finance protocols.
Stablecoin issuers face direct exposure. A stablecoin's stability relies on its reserves and redemption promise. If holders, or their agents, can redeem or rotate stablecoin holdings at scale in seconds, the issuer's liquidity buffer must absorb rapid outflows on a timescale traditional treasury management was not designed for.
The Federal Reserve's recently proposed stablecoin rules, outlined under the GENIUS Act, prioritize reserve quality and redemption mechanisms for precisely this reason. The regulatory framework aims to ensure stablecoin issuers can manage rapid liquidity demands.
DeFi has already shown instances of fast, correlated capital flight. When a protocol experiences an exploit or a stablecoin's peg wobbles, liquidity can exit at speed. This real-world on-chain behavior provides a tangible example of the "agentic run" Slok projects for the broader financial system.
The preference for yield is evident in on-chain data. Blockchains serve as permissionless execution environments where agents can shift funds between stablecoins, which typically offer no native yield, and tokenized funds that provide yield. This continuous movement toward higher-return options highlights active optimization by on-chain capital.
Agentic AI assistants, such as Muse, could funnel household cash into accounts paying 3.3 percent to 5.0 percent, compared to a 0.1 percent national average on checking accounts. This yield differential drives the optimization behavior Slok describes.
