US Banks Lock Out AI Agents Using Customer Credentials, Citing Security Risks

Major US banks are moving to block AI agents that log into customer accounts using user-provided credentials, a practice that violates bank policies and raises serious security concerns. Morgan Stanley, JPMorgan Chase, Bank of America, Wells Fargo, and Citigroup have all begun restricting or entirely blocking external AI agent access, according to a report from Axios cited by The Decoder.

The development marks a significant clash between consumer experimentation with AI automation tools and the strict security requirements of financial institutions. Some customers have been using AI agents to manage account tasks without direct instructions, prompting banks to act.

What banks are doing

Morgan Stanley has already completely blocked AI agents from accessing its platforms. The bank stated it has not observed any malicious or rogue AI agents on its systems, but the preemptive block reflects a conservative approach to an emerging threat vector. JPMorgan, Bank of America, and Citigroup are implementing similar restrictions.

The core issue is that current bank policies require account holders to log in personally. When an AI agent uses a customer’s credentials, it creates an authentication gap. Banks cannot reliably determine whether an agent is acting on explicit user instructions or operating independently.

Most institutions plan to limit any permitted AI agent access to read-only functionality. This would allow agents to view account balances, transaction history, or statements but prevent them from initiating transfers, payments, or any other actions.

How customers are giving AI agents access

Users are enabling AI agents through browser-based tools. Products like Computer Use and open-source agent software such as OpenClaw allow users to delegate browser control to an AI agent. The agent then navigates to the bank’s website, enters the user’s credentials, and performs tasks.

This approach bypasses official bank APIs and formal integrations. Banks have limited ability to block such access on the user’s device, making provider-side restrictions a partial solution. The challenge is similar to distinguishing between a human user and an automated script at the network level.

ChatGPT’s financial service and the read-only model

OpenAI has launched a dedicated ChatGPT service for financial tasks in the US, which operates with read-only access to financial data. This approach aligns with what banks are now demanding: agents can view information but cannot execute transactions.

The OpenAI model provides a structured alternative to ad-hoc credential sharing. It suggests that a viable path forward exists if AI companies and financial institutions cooperate on access boundaries. However, the broader ecosystem of unregulated agent tools remains a concern for banks.

Why this matters for Indian readers and businesses

Indian banks and fintech companies are closely watching these developments. As AI agent adoption grows globally, Indian financial institutions may face similar challenges. The Reserve Bank of India’s existing guidelines on third-party access and data security could influence how Indian banks respond.

For Indian businesses using AI tools for financial management, the US bank restrictions serve as a warning. Relying on credential-sharing with AI agents carries compliance and security risks that regulators and banks may soon address. Indian startups building AI financial assistants should consider read-only architectures and API-based integrations as safer alternatives.

The broader pattern is clear: financial institutions worldwide are likely to follow the US lead. Banks cannot easily distinguish human users from AI agents at the login level, and the security implications of autonomous account access are too significant to ignore.

Datos clave

Bank	Action Taken
Morgan Stanley	Fully blocks AI agents
JPMorgan Chase	Restricting AI agent access
Bank of America	Restricting AI agent access
Citigroup	Restricting AI agent access

Source: The Decoder, “Major US banks are locking out AI agents that log in with customer credentials” (https://the-decoder.com/major-us-banks-are-locking-out-ai-agents-that-log-in-with-customer-credentials/)