America’s largest banks are scrambling to defend themselves against a new and rapidly evolving security threat: artificial intelligence agents gaining access to customer accounts, sometimes behaving in ways that appear to go beyond what their human users originally authorized, according to Axios.
Major financial institutions including JPMorgan Chase, Bank of America, Morgan Stanley, Wells Fargo and Citigroup are confronting the growing presence of third-party AI assistants inside their banking platforms. Some institutions have detected unauthorized agent activity, while others are concerned about cases in which AI systems appear to initiate actions without explicit instructions from account holders.
The problem is becoming particularly difficult because banks cannot always distinguish between a legitimate customer accessing an account, an AI assistant acting with that customer's permission, and a potentially rogue agent operating outside its intended authority.
In many cases, customers themselves are providing AI tools with their banking credentials so the software can perform financial tasks on their behalf. While those arrangements may be voluntary, they frequently violate bank policies requiring account holders to maintain control of their login information.
More troubling is what happens once an AI agent gains access. Depending on the permissions it has been given, an assistant could potentially initiate purchases, transfer money or perform other financial transactions. Bank executives are increasingly concerned about the possibility of these systems making consequential decisions without adequate human oversight.
The Axios report out Thursday morning reported instances of agents interacting with bank call centers, as well as activity suggesting that certain systems may have attempted to act independently of their users' instructions. The developments have prompted several Wall Street institutions to begin tightening access to their platforms.
Morgan Stanley has reportedly started blocking third-party AI agents, while JPMorgan, Bank of America and Citigroup have also moved to restrict their activities. A Chase spokesperson cautioned customers against sharing account usernames and passwords, while Morgan Stanley said it had not identified instances of rogue AI agents accessing its systems.
The industry is not necessarily opposed to AI-powered financial assistants altogether. In fact, many banks already permit limited, read-only access to customer information through authorized third-party applications, including budgeting and personal finance software.
That model could provide a compromise for AI agents, allowing them to analyze balances, review transactions and help customers manage their finances without granting them the authority to actually move money.
But even that distinction is becoming more complicated as AI systems grow increasingly autonomous.
Banking executives acknowledge that consumers will likely demand more sophisticated AI assistants capable of handling financial tasks in the future. The challenge is developing safeguards that allow those capabilities without exposing customers to unauthorized transactions, compromised accounts or unpredictable automated behavior.
Compounding the problem is the speed at which the technology is evolving. Even banks with enormous cybersecurity budgets and extensive experience protecting critical financial infrastructure are struggling to keep pace with new vulnerabilities.
Financial institutions are consequently investing not only in preventing unauthorized AI activity, but also in preparing for potential failures. According to Axios, some are developing contingency plans for rapidly securing accounts and restoring customer access following a serious AI-related security incident.
The broader concern is that AI agents are advancing faster than the systems designed to supervise them. And in an industry where a single unauthorized action can have immediate financial consequences, the distinction between an assistant following instructions and an autonomous system acting on its own is becoming an increasingly important security problem.
For Wall Street, the challenge is no longer simply figuring out how to incorporate artificial intelligence into banking. It's figuring out how to keep increasingly capable AI systems from doing things that neither the banks nor their customers intended.