Bank of America Warns Customers Against Sharing Banking Passwords with AI Assistants

Bank of America Issues Direct Warning on AI Assistant Access

Bank of America sent an email blast titled "Protecting your finances in the AI World" to customers, explicitly warning against sharing banking passwords with AI assistants or any AI-enabled tool that offers to act on their behalf1

1

. The warning comes as customers increasingly hand over login credentials to personal AI assistants like Meta's Muse, Anthropic's Claude, and OpenAI's ChatGPT, which can perform tasks across apps and services2

2

. The email emphasized that customers remain responsible for any activity resulting from shared credentials, effectively placing liability squarely on users who grant access to these tools3

3

.

Growing Trend of Unauthorized Access to Accounts

Major US banks have detected AI agents gaining unauthorized access to customer accounts and acting independently of user instructions, according to an Axios investigation2

2

. This pattern of third-party AI agent access has prompted swift action across the financial sector. JPMorganChase has already blocked AI agents like Muse and Instinct from accessing bank accounts, with CEO Jamie Dimon stating during an earnings call that allowing outside agents inside the company poses too much risk for clients and the institution2

2

. Morgan Stanley, Citi, and Wells Fargo are similarly restricting third-party AI agent access as banks struggle to distinguish between legitimate instructions and unauthorized actions3

3

.

Privacy and Security Risks of Sharing Banking Credentials with AI

Sharing banking passwords with AI assistants exposes customers to multiple security vulnerabilities. A security flaw in an AI assistant could result in credential leaks, giving financially motivated hackers direct access to bank accounts1

1

. AI agents can also make financial decisions without user knowledge, or trigger security protocols that lock customers out of their own accounts when banks flag transactions as suspicious4

4

. The fast-growing agent Muse, which reached over 10 million users in just 40 days since its September launch, requests extensive permissions including contacts, calendar, email messages, login credentials, photos, videos, financial information, health data, and web browsing history2

2

. Muse's privacy policy indicates this data may be used or shared, creating what experts describe as a privacy nightmare for banking customers2

2

.

Agentic Payments Create Regulatory Headaches

The rise of agentic payments, where AI assistants autonomously purchase items at optimal prices or times, has created significant challenges for banks and payment providers1

1

. While the convenience of having an agent purchase out-of-stock items or price drops seems appealing, these transactions are frequently flagged as fraudulent transactions by financial institutions4

4

. Customers have been locked out of accounts following purchases made by AI agents, and some face battles with banks to recover money from unwanted transactions4

4

. Banks, payment providers, and processors are still figuring out how to allow customers access to their accounts using AI tools while maintaining security standards1

1

.

Liability Issues Remain Unresolved

Account agreements at most banks hold customers responsible for anyone they give their login credentials to, creating complex liability issues when AI agents act unexpectedly1

1

. The financial industry hasn't fully resolved who bears responsibility when things go wrong: the customer who granted access, the company behind the AI agent like OpenAI or Anthropic, or the bank that processed the transaction3

3

. While federal protections exist for genuinely unauthorized electronic transfers, the legal landscape becomes murky when customers voluntarily share credentials3

3

. Experts recommend treating AI assistants as strangers with instant power to execute requests, often not in the intended way, with every conversation feeding back to train future models4

4

. Safer alternatives include using AI to analyze spending and build budgets without providing banking passwords, or utilizing bank-authorized read-only connections where supported3

3

.