AI agents could drain $5.4tn from US current accounts, Bank of America warns
Meta's personal AI agent Muse can already read a user's balances, flag wasted subscriptions and point out that money sitting in a current account at 0.1% could be earning 4% or 5% in a money market fund. Bank of America responded with a client note calling the product "a multi-year evolutionary risk" to the banking business model, and Apollo chief economist Torsten Sløk asked whether the industry is seeing the start of an "agentic bank run" — AI agents sweeping cash out of low-paying current accounts into higher-yielding destinations.
The sums at stake are large. Federal Reserve data put about $5.4tn in US current accounts earning close to zero, while the European Commission estimates roughly €10tn of euro-area household money sits in bank accounts paying under 1% a year. If an agent starts comparing yields, identifying better-paid deposits or equally safe sovereign bonds and executing the switch, the cost of funding deposits would rise sharply — and with it pressure on the intermediation margins that underpin traditional banking. JPMorgan has already raised its Meta price target, arguing Muse could become the most-used consumer AI app since ChatGPT. Muse cannot yet move money between accounts on its own, but the plumbing exists: Plaid already connects payment credentials to more than 12,000 US financial institutions.
AI is also testing the banks' defences. Anthropic's September 2026 threat intelligence report said that between December 2025 and August 2026 its team identified and disrupted operations in which state-sponsored groups, financially motivated criminals and lone hacktivists used AI models to automate entire attack chains — from reconnaissance and exploitation of vulnerabilities to data exfiltration, including automatically rewriting malware to evade security detection. Anthropic documented a Chinese group that found more than a dozen potential zero-day vulnerabilities in a single month using AI "agent swarms". For a sector that has digitised its operations and holds millions of customers' data on critical infrastructure, the report implies cyber security spending must scale in step — another cost pressing on margins.
Internally, banks are betting on the same technology to cut costs. Barclays said it will cut 5,000 jobs, about 5% of its global workforce, as a direct result of AI adoption in its processes; chief executive CS Venkatakrishnan said the bank will "employ AI at scale". HSBC, Standard Chartered and Danske Bank have also announced automation-linked job cuts this year. Barclays expects its efficiency measures to generate an additional £1.1bn of savings in 2027. Will Howlett, an analyst at Quilter Cheviot, said the cuts "illustrate how significantly AI and automation could reshape the banking industry in the coming years".
Three forces are converging: agents that bargain on customers' behalf and threaten the cheap deposits that have historically fuelled lending; an attack surface that autonomous AI agents can probe faster than traditional defences can respond; and internal adoption that lifts efficiency while shrinking headcount. Banks that build their own AI assistants to keep deposits inside their ecosystem, invest in advanced cyber defence and manage the workforce transition responsibly will be best placed. The outcome, as with much in this field, is still being written — and AI is moving faster than regulation, organisational adaptation and, possibly, the ability to anticipate its consequences.
What it means for Pakistan: The dynamic described here — AI agents steering savers toward higher-yielding deposits and government securities — maps onto a Pakistani banking system where the State Bank's policy rate has made conventional savings and term deposits far more rewarding than they were during the low-rate years, and where depositors have historically been slow to switch institutions. Pakistani banks are also digitising rapidly under State Bank and SECP oversight, which widens the same cyber-security exposure Anthropic describes; any large-scale adoption of AI assistants by local banks or fintechs would fall under the State Bank's existing technology risk and outsourcing rules, and PVARA's remit does not extend to bank deposits. No Pakistani lender has announced AI-linked job cuts or an agentic deposit product.
