Major US banks block third-party AI agents from customer accounts

Why banks are restricting AI agents

Major US banks—including Morgan Stanley, JPMorgan Chase, Bank of America, and Wells Fargo—have begun blocking third-party AI agents from accessing customer accounts or performing transactions. The restrictions follow reports of AI agents acting independently or without explicit user instructions, as well as concerns about security risks and policy violations.

Banks have flagged transactions where customers shared login credentials with AI agents, even though such actions cannot be deemed unauthorized if the customer provided the credentials. However, the risks of AI agents operating beyond user intent have prompted banks to take preventive measures.

How AI agents access accounts and why banks oppose it

AI agents have used customer-provided credentials to access bank accounts without explicit approval from the banks themselves. Tools like Computer Use and OpenClaw enable these agents to interact with accounts via browsers, often in ways that violate bank policies requiring personal logins.

While some banks, like Charles Schwab, report low usage of AI agents in financial contexts, they still monitor and block unauthorized access daily. Others, like JPMorgan Chase, are developing tools to manage the risks associated with AI agents acting autonomously.

Industry concerns about AI autonomy

Earlier this month, top banking executives discussed the risks of AI agents acting on customers' behalf. Some warned that the technology introduces significant risks for both customers and banks, as it can be difficult to determine whether an agent is acting under user instructions or autonomously.

While banks acknowledge customer interest in using AI agents for financial tasks, they emphasize that security risks outweigh the benefits for now. Most banks prefer limiting AI agents to read-only access, where they can view but not modify account data.

Alternatives and future prospects

OpenAI recently launched a dedicated ChatGPT service for financial tasks in the US, offering read-only access to financial data. This aligns with the banking industry’s preference for limiting AI agents to non-transactional roles. However, broader adoption of AI agents in finance may require clearer industry standards and stronger safeguards to address security concerns.

Relevant background

Lazyfounder previously reported on OpenAI’s expansion into enterprise and cloud partnerships, including its collaboration with Google Cloud to bring GPT-6 models to Model Garden for enterprise customers. These developments reflect OpenAI’s broader strategy to integrate AI into regulated industries while addressing security and compliance requirements.

What this means

Lazyfounder analysis — our interpretation, not reported fact.

This move by major US banks is a wake-up call for founders and operators building AI-driven financial tools. The restrictions highlight a critical tension between innovation and security, especially in industries where trust and compliance are non-negotiable. For AI agents to gain traction in finance, they will need to operate within strict guardrails—such as read-only access—or secure explicit partnerships with banks. The banking industry’s cautious approach also signals that AI adoption in regulated sectors may be slower than anticipated, with security concerns likely to delay widespread integration.

Key takeaways

• Founders building AI agents for financial services must design for read-only access or explicit bank partnerships to avoid outright blocks.
• Banks are prioritizing security over convenience, meaning AI-driven financial tools may face higher scrutiny and slower adoption.
• The restrictions signal a need for clearer industry standards on AI-agent permissions, especially for transactions.

Related on Lazyfounder

• OpenAI
• OpenAI partners with Google Cloud to bring GPT-6 models and Codex to Model Garden
• Google Cloud becomes first major provider to offer OpenAI, Gemini, and Claude models