Banks are blocking AI agents from accounts and governments may follow

The banking crackdown

Major U.S. financial institutions are moving to block personal AI agents from accessing customer accounts and performing transactions. This shift comes as banks detect unvetted AI agents operating inside accounts and attempting to take actions independent of user instructions. Institutions like Morgan Stanley, Bank of America, and JPMorgan Chase are racing to contain these risks.

Citigroup has explicitly stated it does not permit third-party AI agents to log into customer accounts or transact directly on its digital platforms. This policy is designed to safeguard customer information and prevent unauthorized access. The move highlights a growing concern regarding the security of automated assistants that use human login credentials.

Bank of America CEO Brian Moynihan emphasized the need for a human to stay in the loop for money moving out of accounts. He noted that banks are working with agentic providers to ensure consumers do not find money moved by an agent without their knowledge. Currently, many of these banks allow 'read-only' functions, such as viewing balances, while restricting active transactions.

Risks of rogue agents

The primary concerns surrounding AI agents include uncontrolled account access, rogue automated transactions, and unclear legal liability. Experts worry about 'mass deposit flight' and the difficulty of proving whether a payment was actually authorized by a human. When an agent makes a mistake, determining which entity bears the risk remains a significant legal hurdle.

JPMorgan Chase CEO Jamie Dimon warned that allowing outside agents to operate on behalf of customers is too risky. He described the potential threat as 'rogue agents' that could number in the thousands. This stance follows an earlier dispute where the bank moved to charge data middlemen for access to customer data, prioritizing security over ease of access.

The risk is compounded by the fact that many personal AI agents are designed to be 'always-on' background workers. Unlike standard chatbots, these tools can manage files, log into various websites, and execute multi-step tasks. This persistent capability makes them a high-value target for unauthorized use or unexpected autonomous behavior.

Government portal readiness

There is a growing debate over whether government websites should follow the banking industry's lead by blocking AI agents. Government portals handle sensitive tasks such as paying property taxes, renewing driver's licenses, and issuing vital records. Many of these activities require strict identity verification and legal digital signatures that current AI frameworks may not safely handle.

A cybersecurity expert with decades of experience in government portals argues that state and local governments are not yet ready for these agents. Most current government user policies already prohibit third parties from accessing accounts or sharing login credentials. The expert suggests that AI agents should currently be categorized as third parties, making their use a violation of existing terms of use.

While some automation exists for recurring tasks like utility payments, the initial setup and verification are typically established by a human. The expert suggests that government leaders should act now to prohibit AI agents from conducting online business until portals are truly ready. This proactive approach aims to maintain the integrity and trust of public digital services.

The path forward

The landscape of 'agentic commerce' is expanding as companies like Meta hope to collect a percentage of transactions made by AI agents. Testing has shown that about one-third of leading e-commerce and travel websites already block access to these agents, while others challenge or restrict them. This indicates a fragmented environment where different platforms have different levels of readiness for AI interaction.

Legal experts suggest that any future integration of AI agents into government or financial services will require rigorous oversight. Questions remain regarding whether an agent performs a regulated activity, such as providing health guidance or screening applicants. These actions can trigger specific licensing, registration, or conduct rules that AI models may not currently satisfy.

For government agencies, the next steps involve examining current portal policies and terms of use. Decision-makers must determine which outputs require a human review and which actions need explicit approval. Until these security and legal frameworks are established, many experts recommend maintaining a strict 'human-in-the-loop' requirement for all sensitive digital transactions.