The Deposit-Drain Debate: Economists Warn Consumer AI Agents Could Spark the Next Bank Run
A new systemic-risk debate crystallized this week around consumer AI agents and bank deposits. The Times (Chris Dorrell, Oct 2, 2026) reports that the first generation of personal agents — Meta's Muse (launched September, already 3 million weekly users) and OpenAI's Dots (unveiled Tuesday, Sept 29) — has economists worried that agents plugged into bank accounts could sweep cash from low-yield accounts into high-yield fintechs. Apollo chief economist Torsten Slok flagged the risk in a note this week: the average FDIC-insured savings rate is 0.4% versus 3–5% at smaller fintechs. His warning, verbatim: "If every household used AI agents to optimise the return on their cash balances, banks could lose a large share of the cheap deposits they rely on to make loans, which would be a problem for the entire financial system."
The concern is being taken seriously by safety institutions: the UK's AI Security Institute studied agents' impact on banking alongside the Bank of England (study unpublished). Andrew Strait, AISI's former head of societal resilience who led the work, said the problems could be "quite severe": "This is one of many highly plausible and foreseeable pathways to systemic economic harm that requires a high level of co-ordinated action from major developers and banks to address."
The counterargument came from OpenAI's own head of personal finance, Ethan Bloch: "Consumers want a brand name they can trust with their life savings… Chase has $1trn in deposits even though 98 per cent of it could be earning 350 times the yield. Agents don't change any of this." RBC banking analyst Benjamin Toms adds that customers are unlikely to share passwords with agents short-term, but that AI could "structurally compress deposit margins" long term.
Market signal: banking stocks have underperformed since Muse's launch — the S&P 500 Financials index dropped to its lowest level since July last month. Today the risk is hypothetical (no launched agent can move money between bank accounts; Amazon has blocked Muse from purchases outright), but the preconditions map exactly onto the infrastructure being built: agent-permission APIs, agent-native bank accounts (ICBA Sues to Overturn the OCC Trust-Bank Rule Behind Catena's AI-Agent-Native Charter), and headless banking rails (Headless Banking Arrives: Paywhere's Programmable AI Storefront Puts Banks Inside ChatGPT and Claude). For landscape-mapping purposes: bank agent-permission policy is becoming a strategic decision — JPMorgan's cautionary stance (JPMorgan Payments Charts Cautious Strategy on Agentic Commerce, Flags Liability Gaps) now has a macroeconomic rationale attached to it.