U.S. Banks Detect AI Agents in Customer Accounts and Restrict Their Access
Key Facts
Major U.S. banks have detected third-party AI agents operating inside customer accounts, with some appearing to take steps independently of user instructions, Axios sources said. The report attributed the information to people familiar with the matter and did not report any announcement by Charles Schwab that it had detected similar cases. It also said Morgan Stanley, Bank of America, JPMorgan and Wells Fargo were among the banks seeking to contain agent-related risks. The issue centers on distinguishing between a customer using an agent with their consent and a bank allowing that access. The report said some banks were aware of unauthorized access, without establishing that their protocols had been technically breached.
The access cases often involved an agent or assistant using login credentials provided by the customer and with the customer’s permission, the report said. It quoted a source describing the cases as "situations where an agent or AI assistant accesses a person's account with the user's consent and login credentials." But customer consent does not necessarily mean the bank authorized the agent to use the account. The report said some banks require customers to log in themselves, making the sharing of login credentials problematic even when a customer asks an agent to access the account. The report therefore described the agents’ activity as unauthorized access from the bank’s perspective, not as evidence in itself that technical protections had been bypassed.
Banks’ responses range from blocking external agents to limiting what they can do inside accounts. Morgan Stanley had begun blocking third-party agents from its platforms, while JPMorgan, Bank of America and Citi were restricting their access, Axios reported. Morgan Stanley, however, said through a spokesperson that the bank had not detected rogue AI agents accessing its platforms. That describes the bank’s position on detecting rogue agents, while the report separately said the bank had begun blocking external agents. Most banks are moving toward allowing read-only access, which lets agents view financial information without taking action, according to the report.
Banks are struggling to determine whether an agent’s activity inside an account was directed by the user, exceeded the scope of its authorization or took place without permission. Some agents contacted customer service centers, while others appeared to take steps independently of user instructions, the report said. In the broader context of protecting brokerage accounts, FINRA warned in its 2026 oversight report that agents could act autonomously without human verification or go beyond the scope intended by the user. It recommended considering access and data monitoring, tracking agents’ decisions and setting controls on their behavior. It also cited practices for reviewing suspicious activity, including transfer requests to previously unused third parties and logins from unfamiliar browsers or locations.
FINRA sets different compliance deadlines for amendments to Regulation S-P on protecting customer information: larger entities had to comply by 3 December 2025, and smaller entities by 3 June 2026. The regulator said broker-dealers must establish policies and controls to protect that information. On 30 September 2026, FINRA said it had seen an increase in fraud schemes targeting investors, particularly older people, and urging them to transfer money from brokerage and other financial accounts to external accounts. The report did not directly link those schemes to AI-agent activity, but cited them in the context of account and transfer security. Some banks are also preparing for post-incident scenarios, including plans to secure or quickly restore customer access, Axios reported.