Citi and Bank of America restrict AI agents from customer accounts

Citigroup and Bank of America have joined other US banks in restricting AI agents from accessing client accounts.[1] One concern is that an agent could move idle cash to seek higher returns, draining deposits from the bank where the money is held.[1] The reported restrictions matter to products designed to inspect or act on a user’s finances: many compelling use cases depend on an agent being able to see and act on bank accounts.[2] The evidence does not specify the precise access controls Citi and Bank of America are applying.[1]

Citi CEO Jane Fraser said autonomous AI agents need rigorous identity verification and authentication before they can execute transactions.[2] She said those requirements apply whether an agent acts for a customer or for the bank.[2] Citi’s ARC platform is an internal system for monitoring and governing AI agents, and Fraser said agents cannot be created at Citi without going through it.[2] ARC was introduced in 2026 and is mandatory for overseeing Citi’s internal AI agents.[2] Fraser said she was confident Citi could avoid risks associated with uncontrolled agents.[2]

Citi’s emphasis on internal governance is distinct from JPMorgan CEO Dimon’s stated total ban on external AI agents accessing customer accounts.[2] Dimon cited the danger of “rogue agents.”[2] At Charles Schwab, CEO Rick Wurster said the company actively blocks unauthorized agents attempting to move clients’ money, including Meta’s Muse.[3] Wurster also said Schwab does not want customers to give Muse their passwords and let it act for them, citing uncertainty about control and liability.[3] These examples show different stated approaches, not a single uniform bank policy.

Restricting account access may complicate products whose agents need to view and act on a user’s finances.[2] If the largest banks wall off customer accounts, those products lose access to a large part of the market.[2] Governance systems such as ARC also require staff, systems and ongoing monitoring.[2] Citi has been gradually rolling out governed AI systems internally since at least mid-2025, with an emphasis on compliance and oversight.[2] For teams building financial agents, the evidence points to a design tension: useful account actions depend on access, while banks are emphasizing authentication, control and governance.