📰 Key Takeaways

Bank of America recently sent customers an email titled “Protecting your finances in the AI World,” explicitly urging them not to give their online banking username and password to anyone else or any outside service, including AI tools or AI agents that claim they can operate the account for them. The background is that as personal AI assistants become more common, more and more users are handing their login details directly to AI. The report points to three major risks. First, the AI could spend money without the user knowing. Second, the bank’s fraud systems might read the activity as someone hijacking the account and lock the user out. Third, if the AI assistant itself has a security hole, the banking credentials could leak, giving profit-driven hackers a way into the account. Banks, payment companies, and processors are still figuring out how to let customers access accounts through AI tools, such as “agentic payments,” where an AI agent automatically shops for you at the best price or time. They’re also reportedly dealing with unauthorized AI agents accessing customer funds. A Bank of America spokesperson confirmed the email is real. The email also reminds customers who’s responsible: most banks’ account agreements say customers are liable for anything done by someone who has their login information, and the email notes that if you still choose to share your credentials, the resulting activity is your own responsibility.

💬 JudyAI Lab Take

Bank of America recently emailed customers to remind them not to hand their online banking username and password to any AI tool or AI agent. This is worth paying attention to because it’s the first time the financial industry has publicly said “stop” to giving login details to AI.

This case shows that once AI agents move into everyday life, the first bottleneck is authorization, not capability. The original report points out that once users hand over their credentials, the AI could spend money without them knowing, the bank’s fraud systems could mistake it for account theft and lock the account, and if the AI assistant itself has a vulnerability, the credentials could leak to hackers. Meanwhile, banks and payment companies are still figuring out things like “agentic payments,” where AI shops on your behalf. The email also stresses that most account agreements make the customer responsible for everything done by anyone who has their login information. For AI builders, this means you can’t treat “getting the user’s password” as a shortcut when designing an agent. Permission scope, auditable logs, and accountability all need to be figured out from the very start of the product.

If you’re building or using an AI tool that touches money, first check whether it asks you to type in your password directly, and see if there’s a narrower-permission way to authorize it.

📅 Source Info

🔗 Further Reading