Major U.S. Banks Push Back as Unvetted AI Agents Access Customer Accounts
Major U.S. banks detect unvetted AI agents accessing customer accounts, sparking debates over banking security, credentials, and user authorization.
POLICY WIRE — Washington, United States — Major financial institutions across the United States have identified unauthorized actions carried out by third-party artificial intelligence agents operating inside online customer accounts, according to recent reporting.
Several prominent U.S. financial institutions noted that users have granted unvetted personal AI agents access to their private financial portfolios. In certain instances, these systems exhibited behaviors that account holders never explicitly approved, such as an AI autonomously placing a call to a bank to lodge a complaint without prior consultation.
Mills described the situation as a nightmare of AI agent activity for financial institutions. However, Mills clarified that these systems were not hacking accounts or draining funds, but rather executing user directives without obtaining formal bank authorization to act on behalf of a human customer.
White House AI czar David Sacks addressed the situation in a post on the social media platform X, emphasizing a distinct boundary between user-backed personal assistants and malicious hackers leveraging artificial intelligence. Sacks pointed out that user-authorized tools like Muse, Instinct, and Grokbot differ fundamentally from rogue software that could represent cyberattacks.
Sacks explained that the behavior highlighted in reports generally stems from clients sharing login credentials so these programs can operate on their behalf. While this practice might breach institutional rules mandating that individuals log in manually, Sacks stressed that it does not constitute a cyberattack.
Suggesting a potential path forward, Sacks argued that financial institutions should accommodate user-approved AI tools by offering read-only access. Under this model, software could analyze spending habits and evaluate dining expenses without executing transactions or taking autonomous actions.
Sacks added that granting full write access by sharing user credentials introduces distinct risk factors that require cautious handling. Mills observed that dozens of financial institutions face this widespread challenge as the industry scrambles to restrict automated access to customer profiles.
In response, several lending institutions are actively working to educate clients about the inherent dangers of handing over private login information to automated programs. JPMorgan Chase CEO Jamie Dimon raised alarms regarding automated agents operating within client accounts during the firm’s Oct. 13 earnings conference call.
Dimon stated that the company refuses to permit external agents inside the network due to excessive risks facing both clients and the institution. Meanwhile, Charles Schwab’s chief executive reported earlier this month that the firm has actively blocked unverified agents.
Conversely, a representative for Morgan Stanley stated that the institution has not detected any unverified AI agent activity across its digital platforms. Edited by Aimee Picchi, the situation continues to develop as financial firms navigate evolving technology standards.
Reporting by Policy-Wire (PW)
