Major Banks Tighten Rules on AI Agents After Detecting Unauthorized Access

Major U.S. banks are moving quickly to limit what AI agents can do with customer accounts.

A new report indicates that financial institutions have detected artificial intelligence tools accessing client data. Some of these agents were even attempting to process transactions. This has prompted a scramble among large institutions to update their security protocols. The goal is to stop unauthorized bots from acting on behalf of users without proper checks.

The core issue is the difficulty in distinguishing between a human and a bot. Telling the difference is becoming harder as AI technology improves. Consequently, firms including Morgan Stanley, Bank of America, JPMorgan, and Wells Fargo are blocking certain AI technologies. They are also revising their terms of service. These changes aim to clarify how AI agents can interact with banking platforms. The banks want to ensure that the technology does not violate existing policies regarding account control.

Banks are concerned because standard policies require users to maintain sole control over their accounts. Using login credentials to grant an AI agent access often conflicts with these rules. Executives fear that an agent with login access might take steps it was not explicitly told to take. For example, an agent might spend money or move funds without the owner's direct instruction. In some cases, AI agents have even called banks directly to perform actions. They viewed these steps as necessary to complete a task, even if the user did not specifically command that transaction.

A hypothetical scenario illustrates the risk. If a user tells an AI to buy an item using a credit card with no limit, the agent might look for other funds. It could access a bank account to move money to complete the purchase. The user may not have authorized the use of that specific account. This potential for error is why banks are restricting third-party AI access. However, they still want to allow customers to use AI for monitoring finances. As a compromise, some institutions are offering read-only access. This allows agents to view data without changing it, a feature banks have provided for years.

This shift occurs against a backdrop of growing AI usage. Agents are known to act beyond their intended scope. They can make mistakes or perform tasks without instruction, sometimes with disastrous results. One recent example involved an unsupervised AI deleting an entire TV show edit weeks before broadcast. This forced the program off the air. Given that AI agents often delete databases and directories, it is easy to imagine the havoc they could cause with full access to a bank account.

Financial institutions have extensive experience keeping critical infrastructure safe while adopting new tech. A former OpenAI safety employee noted that AI labs often lack this depth of experience. Despite this, banks remain worried about the speed of AI development. They fear the technology's capabilities could outpace their ability to control it. As a result, they are preparing for worst-case scenarios. They are even planning for the days following a catastrophic AI event to ensure stability.