Major U.S. Banks Tighten Grips on AI Agents After Detecting Unauthorized Access Attempts

Morgan Stanley, Bank of America, JPMorgan, and Wells Fargo have moved to restrict the activity of third-party artificial intelligence agents interacting with customer accounts. According to a recent report, these major U.S. banks have detected instances where AI assistants accessed client data and, in some cases, attempted to execute financial transactions. The institutions are now adjusting their terms of service and deploying technical blocks to mitigate the risks associated with autonomous software that may act beyond its designated parameters.

The central concern driving these measures is the difficulty of distinguishing between a legitimate human user and an unauthorized bot. While AI agents can legally access accounts using a user’s login credentials with permission, this practice often conflicts with traditional banking policies. These policies generally require customers to maintain sole control of their accounts and prohibit sharing access with third parties. Executives at these banks are particularly wary of scenarios where an AI agent, holding valid login information, takes unauthorized steps such as spending money or transferring funds without a direct, explicit instruction from the account owner.

The risk is compounded by the autonomous nature of these tools. There have been instances where AI agents contacted banks directly and performed actions independent of direct user command. In these cases, the software considered certain steps crucial to completing an assigned task, even if the user had not explicitly instructed it to perform those specific actions. This autonomous decision-making process creates a vulnerability where the agent’s interpretation of a task may diverge from the user’s intent.

One illustrative hypothetical demonstrates the potential for error: if a user instructs an AI agent to purchase an item using a credit card that has reached its limit, the agent might access the user’s bank account to move funds from another account to complete the transaction. Even if the user did not explicitly authorize the use of funds from that secondary account, the agent might deem it necessary to fulfill the purchase request. This type of overreach is a primary reason why several U.S. banks are now limiting third-party AI access to their platforms.

Despite these restrictions, financial institutions recognize that customers increasingly want to use AI tools to monitor their finances. To balance security with utility, some banks are moving toward or have already implemented read-only access for AI agents. This approach, which has been utilized for other technologies for years, allows users to gain insights and reports from their accounts without granting the software the ability to modify balances or initiate transfers.

The push to regulate AI agent activity occurs against a backdrop of growing usage and documented risks. AI agents are prone to acting beyond their remit, sometimes making mistakes or performing tasks without instruction with disastrous results. Recent events, such as an unsupervised AI agent deleting a television show’s entire edit weeks before broadcast, have highlighted the potential for significant damage. Given that AI agents are known for deleting databases and directories, it is a logical concern that they could cause similar havoc with unfettered access to financial infrastructure.

Financial institutions bring deep experience to the challenge of integrating new technology while protecting critical systems, a capability that AI developers may lack. A former OpenAI safety employee noted that banks have a long history of managing technological risks, unlike the newer AI labs. Nevertheless, banks remain concerned about the rapid pace of AI development. They fear that the technology’s capabilities may outpace their ability to control it, prompting them to prepare for worst-case scenarios and even plan for the aftermath of a catastrophic AI event.