Leading US financial institutions, including Morgan Stanley, JPMorgan Chase and Charles Schwab, have begun restricting unauthorised third-party artificial intelligence (AI) agents from accessing customer accounts and conducting financial transactions. The measures reflect growing concerns over autonomous AI systems operating within banking environments, potentially performing activities beyond customer instructions and exposing financial institutions to cybersecurity, operational and liability risks.
According to a Reuters report published on 22 October 2026, several major US banks have identified instances of third-party AI agents attempting to access customer accounts. Some agents reportedly operated using login credentials voluntarily shared by customers, creating uncertainty over authentication, consent and the legitimacy of automated financial activities.
The issue has become particularly significant as AI agents evolve beyond conventional conversational tools to perform tasks independently, including accessing financial information, interacting with digital platforms and potentially initiating transactions. Banking executives are concerned that these systems may execute actions that exceed their authorised scope or expose customers to unintended financial consequences.
Morgan Stanley has reportedly implemented restrictions preventing external AI agents from accessing its platforms. However, sources familiar with the institution’s operations indicated that it had not identified instances of rogue or unauthorised AI activity within its systems.
JPMorgan Chase is also developing safeguards to manage risks associated with autonomous AI agents. Chief Executive Officer Jamie Dimon has emphasised the importance of controlling external AI access to protect customers and the institution from potential security and operational vulnerabilities.
Similarly, Charles Schwab has implemented measures to block unauthorised AI agents. Chief Executive Officer Richard Wurster indicated that the institution actively monitors attempted access by such systems, emphasising the need to retain control over authentication procedures and the customer experience.
The developments highlight an emerging challenge in financial services: distinguishing legitimate customer-authorised automation from potentially unsafe or unauthorised AI activity. Even where customers voluntarily provide their account credentials, financial institutions may face difficulties determining whether subsequent actions are consistent with customer consent and established security requirements.
The issue also raises questions about accountability when autonomous AI systems initiate transactions, disclose sensitive financial information or make decisions that result in customer losses. Traditional authentication and fraud prevention frameworks may require further adaptation to address risks arising from non-human digital agents operating across interconnected financial platforms.
Financial institutions are increasingly examining controlled access arrangements, including read-only permissions, restricted transaction capabilities, stronger identity verification and explicit authorisation mechanisms. Such controls could allow customers to benefit from AI-enabled financial services while reducing exposure to unauthorised transactions and third-party technology risks.
The situation represents an important development in AI governance, as banks seek to balance technological innovation with customer protection, operational resilience and regulatory compliance. It also reinforces the need to incorporate agentic AI risks into enterprise risk management, cybersecurity governance and third-party risk management frameworks.
RMAI Risk Perspective: The restrictions imposed by major US financial institutions highlight the emergence of autonomous AI agents as a distinct category of technology and operational risk. Banks should evaluate whether existing third-party risk management frameworks adequately address AI-agent authentication, customer consent, transaction authorisation, activity monitoring and accountability. A critical question for risk professionals is whether financial institutions can establish effective controls that distinguish customer-authorised AI assistance from autonomous actions that exceed permitted boundaries.
