Apollo Chief Economist Torsten Slok: AI Agents May Trigger Bank Runs
Apollo Global Management Chief Economist Torsten Slok warned in the Daily Spark that AI assistants like Muse could automatically transfer household cash from an average national checking account of about 0.1% into accounts yielding 3.3% to 5.0%, creating a "bank run" driven by agents.
He stated that if every household uses agents to optimize cash returns, banks will lose the cheap deposits they rely on for lending, which could impact the entire financial system. Apollo listed 11 fintechs and online accounts with interest rates between 3.3% and 5%; Adelfi around 5%, SoFi around 4.5%. The FDIC national average is about 0.4% for savings and 0.1% for checking. A $10,000 balance at 0.1% yields about $10 annually, while at 5% it yields about $500.
The core of bank profits is low-interest deposits and high-interest loans. Agents turn "lazy deposits" into comparable, transferable funds. Meta's Muse, launched in September, connects accounts via Plaid and can read balance, transaction, and investment information; however, it does not explicitly state that it can automatically transfer funds between banks. The warning is about the scale effect once capabilities are complete, not about the runs that have already occurred.
Similar mechanisms existed in the pre-AI era: Yu'ebao swept idle funds in Alipay into money market funds, and Europe's Raisin matched deposits to higher-interest banks. The difference is that agents execute price comparisons and transfers in the background continuously, no longer limited by the attention of working individuals.
In market mechanics, buyers are high-interest online accounts and money market funds, while sellers are traditional banks relying on 0.1% checking deposits. Funds are flowing from core deposits to the liability side of fintechs. Beneficiaries include high-interest receivers like SoFi and Wealthfront, while lenders relying on lazy deposits face pressure on their net interest margins. The event is driven by this macro note, not by systemic withdrawals that have already occurred.
Source: Public Information
ABAB AI Insight
Slok has long written briefs on interest rates and financial stability for Apollo, this time framing agents as solvents of deposit franchises. A significant portion of bank valuations is based on "customer inertia"; a Harvard study estimated this inertia to account for over half the value of deposit franchises. Agents do not require a run mentality; as long as the interest rate differential exceeds transfer friction, they can execute overnight what Yu'ebao did in China in 2013.
The capital path is the redistribution of interest margins. Once cheap core deposits are swept away, banks will either raise interest rates to attract liabilities or shrink credit. Resources will flow to fintechs that can connect to Plaid and open high-interest accounts, as well as platforms selling agent access. The motivation is to remind credit investors that the assumption of deposit stability will no longer hold after agents become widespread. Apollo itself is simultaneously raising large-scale financing for AI infrastructure, and this note also serves to mark risks for its credit portfolio.
Similar cases include Yu'ebao's impact on bank demand deposits, the accelerated withdrawals from Silicon Valley Bank in 2023 due to Twitter, and the chronic draining of community banks by high-yield savings apps. The industry is in a control period for deposit pricing: agents and high-interest accounts are expanding, while traditional checking accounts still rely on inertia, and the transformation is incomplete.
The structural judgment pertains to the transfer of pricing power. The pricing power of cash deposits is shifting from bank licenses and branch habits to agents that can compare and execute 24/7. The mechanism is: when the cost of migration approaches zero, 0.1% and 5% cannot coexist on the same household balance sheet.
ABAB News · Law of Cognition
- Where lazy deposits disappear, bank franchises begin to discount.
- Runs do not require queues; agents can complete them overnight.
- Interest margins exist because humans are lazier than machines.