Five banks have moved to block third-party AI agents from operating inside customer accounts and making transactions. Given the risks and uncertainties surrounding agentic AI, those banks’ response to AI agents having accessed their customers’ accounts is understandable. Banks are right to be concerned about fraud, unauthorized transactions, privacy risks, liability, and regulatory accountability. AI agents can act with autonomy, operate in real time, and potentially amplify mistakes or malicious activity. But while blocking AI agents may be a prudent short-term fix, it is not a viable long-term strategy. 

Today banks can be cautious because customer demand for fully autonomous financial agents remains limited. Our research shows that consumers are far more comfortable using AI for information gathering, recommendations, and research than they are delegating money movement or making financial decisions. And many major banks and investment firms allow third-party agents to have limited, read-only access through approved open banking providers (such as Plaid). But half (50%) of US and Canadian consumers we surveyed don’t trust AI with sensitive financial information, and 40% have concerns about data security and confidentiality. Trust in autonomous financial agents remains low, and most customers still want to stay in control. 

That won’t last forever. 

As AI assistants become more capable and more deeply embedded in consumers’ daily lives, trust will grow, and customers will increasingly expect them to help compare products, monitor accounts, prepare applications, and eventually initiate transactions. When that happens, banks that simply block all agent access risk creating a different problem: absence (or more accurately put, “abstention”). Customers may come to rely on AI assistants as their primary interface for discovering, evaluating, and managing financial products. If banks are absent from those interactions, they risk relegating themselves to commodity product providers behind AI platforms that own the customer relationship. The banks that refuse to engage with trusted agents could find themselves excluded from the very experiences customers prefer. 

Build A Controlled Path Forward 

The answer is not to move from blocking everything to allowing everything. It is to move from binary decisions to graduated trust. Banks need a framework for determining which agents can access what, under what circumstances, and with what permissions. That means authenticating both the customer and the agent, verifying an agent’s provenance, assigning unique identities to agents, and restricting access to only the permissions necessary for a given task. Unknown or untrusted agents can remain blocked. Trusted customer-authorized agents can be granted tightly controlled access. 

Customer control must remain central. Permissions should be explicit, revocable, time-limited, and tied to clearly defined actions. High-risk activities such as payments and money movement should initially require customer approval, while lower-risk activities such as account research, alerts, monitoring, and form preparation can serve as early proving grounds. Banks must also maintain continuous oversight. Agent activity should be monitored, catalogued, assessed for risk, rate-limited, and subject to machine-aided but human-driven controls that can stop suspicious activity before it causes harm. 

The institutions that succeed in the age of agentic AI will not be those that block agents indefinitely. They will be those that learn how to distinguish trusted, customer-authorized agents from malicious actors. The current wave of blocking reflects a banking industry that is not yet ready for external agents. The next phase will belong to banks that build the infrastructure, controls, and trust frameworks necessary to welcome them safely. 

Want to dig deeper? Forrester clients can schedule a guidance session. Not a client? Let’s talk about how we can help. 

Share