Major US financial institutions including JPMorgan, Bank of America, Wells Fargo, Citi, and Morgan Stanley are responding to unauthorized AI agents operating inside customer accounts. According to an Axios report, these autonomous software tools have accessed accounts using customer credentials and even contacted bank call centers, though no confirmed banking losses have been documented to date.
Banking Giants Clamp Down on Automated Access
In response to automated login activity, Morgan Stanley has reportedly begun blocking third-party AI agents, while JPMorgan, Bank of America, and Citi are implementing strict access restrictions. Several banks are pivoting toward read-only access models, allowing AI tools to check balances without authorization to move funds. During JPMorgan's October 13 earnings call, CFO Jeremy Barnum issued a clear warning regarding customer accountability, stating that if customers voluntarily share credentials with agents that perform unwanted actions, "that's going to be on them."
Despite bank warnings, existing US regulations regarding unauthorized electronic transfers leave customer liability ambiguous when automated tools exceed their intended permissions. Industry analyst Linus Ekenstam noted that autonomous agents searching for optimized deposit yields could potentially disrupt bank liquidity, threatening roughly $500 billion in US bank equity value if automated transfers scale without friction.
Key Takeaways
- JPMorgan, Bank of America, Citi, and Morgan Stanley are blocking or restricting AI agents from moving money inside customer accounts.
- JPMorgan CFO Jeremy Barnum warned on October 13 that customers sharing login credentials with third-party software bear liability for unwanted transactions.
- Crypto exploits detailed by BeInCrypto show $281,200 in combined AI-linked transfers across two separate 2025 security incidents.
- Research shows 14 of 19 reviewed AI products operate within human-approved limits, while 5 of 7 trading platforms explicitly prohibit automated fund withdrawals.
Crypto Precedents Reveal Vulnerability Risks
While traditional banks have avoided confirmed financial losses so far, the digital asset ecosystem highlights the risks of granting autonomous software direct financial controls. As documented in tracking reports on AI agent payment activity, attackers previously exploited automated setups to drain $281,200 in crypto before assets were recovered. In March 2025, an attacker compromised AIXBT's dashboard to execute an unauthorized $106,200 transfer. Two months later in May 2025, a malicious prompt processed through Grok caused Bankr to send $175,000 in tokens, though Bankr subsequently recovered 80% of those funds.
BeInCrypto research indicates that 14 of 19 reviewed AI products and workflows operate within boundaries pre-approved by humans, while 5 of 7 financial trading venues outright ban AI agents from processing withdrawals.
Why It Matters
The clash between bank security protocols and autonomous AI agents represents an inevitable friction point as consumer financial management becomes increasingly automated. While banks are rushing to implement read-only boundaries to protect deposit stability, regulatory uncertainty surrounding electronic transfer liability leaves consumers exposed if autonomous agents misinterpret instructions. Until legal frameworks define explicit boundaries for delegated AI authority, credential-sharing remains a major liability risk for retail bank customers.
