AI Agents Could Trigger 'Agentic Bank Runs' by Auto-Sweeping Deposits to High-Yield Accounts

Retail bank executives have long modeled consumer deposits as a sticky, low-cost funding source, relying on the friction of manual transfers to keep cash in checking accounts that pay near-zero interest. That foundational assumption is now colliding with the deployment of autonomous artificial intelligence agents. In a September 27 note, Apollo Global Management Chief Economist Torsten Sløk warned that AI assistants could soon automatically sweep household balances into high-yield fintech accounts, triggering what he termed an "agentic bank run." By eliminating the inertia that keeps deposits in place, these agents threaten to drain the cheap funding that traditional banks rely on to issue loans, creating a new vector of systemic risk for the financial sector.[1][4]

The economic incentive driving this potential capital flight is a stark yield gap. The Federal Deposit Insurance Corporation reports the national average yield for a checking account is currently 0.1%, while high-yield savings accounts and fintech platforms offer between 3.3% and 5.0%. On a $10,000 balance, a traditional checking account generates roughly $10 in annual interest, compared to $500 at a 5.0% rate. For decades, banks have captured that $490 difference as net interest margin, betting that the administrative hassle of opening new accounts and moving funds would deter most retail customers from chasing better returns.[2][4]

Agentic AI removes that friction entirely, fundamentally altering the behavioral economics of retail banking. Unlike earlier generations of chatbots that merely answered financial queries or provided spending summaries, modern AI agents are designed to execute complex workflows across multiple platforms on a user's behalf. "If every household used AI agents to optimize the return on their cash balances, banks could lose a large share of the cheap deposits they rely on to make loans," Sløk wrote in his analysis. He emphasized that this dynamic poses a problem for the entire financial system, not just individual institutions, because the banking sector's lending capacity is structurally dependent on underpaying for retail liquidity.[1][3][4]

The infrastructure required for this shift is already entering the consumer market. On September 8, 2026, Meta launched its Muse AI assistant, integrating it with the financial data network Plaid. Plaid's application programming interfaces connect to more than 12,000 United States financial institutions, allowing Muse to read balances, track transactions, and monitor investment portfolios. While neither Meta nor Plaid has explicitly enabled Muse to autonomously sweep funds between banks yet, the underlying connectivity exists, and JPMorgan analysts recently projected that Muse could become the most widely adopted consumer AI application since ChatGPT.[1][2]

Consumer readiness for automated financial management is accelerating rapidly as the technology becomes more accessible. A September 2026 report from EY found that 70% of younger consumers are already willing to use AI agents as personal financial assistants, signaling a generational shift in how banking relationships are managed. The consulting firm warned that as deposits become programmable, balances will move instantly across providers, fundamentally altering the operating economics of retail banking. "As AI agents become the primary decision layer, operating economics and risk exposure can change faster than traditional models can adapt," the EY researchers noted.

Market analysts point out that yield-chasing behavior is already highly visible among sophisticated retail investors, even without the catalyst of AI automation. Chartered Financial Analyst Mike Zaccardi noted that he currently routes his cash into BOXX, an exchange-traded fund designed to mirror short-term Treasury bill returns, rather than leaving it idle in a traditional checking account. Zaccardi echoed Sløk's concerns regarding the broader market implications, stating that if AI assistants begin auto-sweeping cash into 5% yields at scale, banks will face a systemic crunch as their foundational deposit bases evaporate overnight.[1][2]

The mechanics of an agentic bank run differ fundamentally from historical financial panics. A traditional run is driven by fear, occurring when depositors lose confidence in an institution's solvency and rush to withdraw funds before reserves are depleted. An agentic run, by contrast, is driven purely by mathematical optimization. The software does not panic; it simply executes its mandate to maximize yield continuously and without hesitation. If millions of AI agents act on the same interest-rate signal simultaneously, the resulting outflow would resemble a coordinated, high-speed drain rather than a chaotic crowd.[4][5]

This dynamic is already forcing a regulatory debate in Washington. Traditional banks are lobbying for legislation that would level the playing field against cryptocurrency exchanges and fintech platforms, which currently offer yields of 5% to 8% on stablecoin holdings. The tension surfaced during the September 15 Senate procedural vote on the Clarity Act, a bill aimed at establishing a federal framework for stablecoin issuers. Banks argue that without regulatory parity, they are structurally disadvantaged in competing for the programmable liquidity that AI agents will inevitably seek.[1][2]

The immediate constraint preventing an agentic bank run today is product design, rather than any underlying limitation in the technology itself. Meta and Plaid currently restrict Muse to read-only access for financial data, deliberately holding back from enabling full autonomous money movement across different institutions. The structural vulnerability, however, is already firmly priced into the market: the 4.9 percentage point gap between traditional checking accounts and fintech yields exists, and the sophisticated software capable of bridging that divide is now live on millions of consumer devices.[1][4]