Bank AI restrictions are tightening around third-party software that accesses customer accounts using shared login credentials. As of October 22, 2026, JPMorgan Chase, Bank of America and Citigroup had restricted outside agents, highlighting the difference between a customer granting software access and a bank accepting that software’s authority to act.
Key Takeaways
- The restrictions target uncontrolled third-party account access, not banks’ internal use of AI.
- Valid login credentials do not establish customer authorization for every action an agent attempts.
- Widespread unauthorized transfers, quantified customer losses and large-scale AI-driven deposit migration have not been established.
Table of Contents
- Why bank AI restrictions focus on authorization
- Controlled connections offer an alternative to shared passwords
- Deposit competition is a separate, longer-term risk
- Conclusion
Why bank AI restrictions focus on authorization
AI assistants can navigate websites, enter passwords and complete multistep tasks. When they use a customer’s banking credentials, their activity can be difficult to distinguish from an ordinary customer session. The harder question is whether each requested action falls within the customer’s instructions.
JPMorgan says it does not permit third parties to log into customer accounts using shared banking credentials. During its October 13, 2026, earnings call, Chief Executive Jamie Dimon warned about the risks of outside agents: “You don’t want rogue agents, and there could be thousands of them.” He also cautioned customers about potential liability when granting external software access.
The scope of the problem remains uncertain. Morgan Stanley said it had not observed rogue AI agents accessing its platforms. The available information does not establish widespread unauthorized fund transfers or quantify customer losses, so the immediate issue is account security and responsibility rather than a demonstrated industry-wide financial hit.
| Item | Details |
|---|---|
| Account selection | Chase allows customers to choose which accounts to link to an app. |
| Revocable access | Customers can review and remove app access through Chase’s linked-app controls. |
| Additional data requests | Chase says it requires customer approval before providing additional data for new app features. |
Controlled connections offer an alternative to shared passwords
Blocking credential-based access does not mean rejecting all third-party financial services. Chase’s secure account-linking process lets customers authenticate directly with the bank rather than disclose their username and password to an outside app. That existing framework explains how controlled data sharing differs from unrestricted agent access; it does not establish the scope of the latest AI restrictions.
A key distinction is permission to retrieve information versus authority to move money. Read-only connections can support balance checks and transaction reviews, while payments may require additional authentication, transaction-specific approval or spending limits. These are possible safeguards, not an announced industry-wide standard.
Deposit competition is a separate, longer-term risk
AI financial assistants could eventually compare deposit rates, identify idle balances and initiate transfers to higher-yielding accounts. If customers authorize those actions and suitable payment connections are available, banks could face less stable low-cost deposits and pressure on funding costs.
On JPMorgan’s third-quarter earnings call, Chief Financial Officer Jeremy Barnum identified yield-seeking deposit flows and fintech competition as established pressures that personal AI agents could combine. That is a potential competitive development, not evidence that automated deposit migration is already happening at scale.
The distinction matters: account-access controls address immediate security and consent questions, while any effect on deposit costs depends on adoption and the permissions customers give agents.
Conclusion
Bank AI restrictions currently draw a boundary around uncontrolled account access, rather than all automated financial services. The practical checkpoint for customers is whether an app uses a bank-approved connection and what permissions it receives; the available information does not establish a timetable for broader agent-access standards.
Read More News
Disclaimer
This FinanceVelix article provides general information, not personalized financial, investment, legal, or tax advice. Market prices, interest rates and economic conditions can change; investments carry a risk of loss. Facts reflect the material available when published and may change. Check important information independently and consult a qualified professional when needed. Mentioning an asset, company, or service is not an endorsement.
