Smashing Security podcast #491: How AI agents are spooking the banks
Bank of America has sent an email to its customers urging them to stop sharing their banking credentials with AI assistants, warning that users remain liable for any resulting financial activity.
The largest bank in the United States has issued a direct warning to its customers about the dangers of sharing login details with artificial intelligence tools.
In an email titled "Protecting Your Finances in the AI World," Bank of America told clients to stop handing over usernames and passwords to AI-enabled agents.
The bank emphasized that users are responsible for any financial activity resulting from the unauthorized sharing of their credentials.
Liability shift
The email serves as a stark reminder that the responsibility for fraud or unauthorized transactions lies with the account holder, not the bank.
Customers are advised to resolve any issues directly with the AI assistant that performed the action, leaving the financial institution to wash its hands of the liability.
This approach effectively shifts the burden of proof and resolution onto the consumer, who must now navigate disputes with software rather than human staff.
Rising use of AI agents
Despite the warnings, consumers are increasingly granting AI agents access to their personal and financial data to automate daily tasks.
A recent episode of the New York Times podcast "The Daily" featured a reporter who turned over highly personal details, including banking information, to Meta's new AI agent called Muse.
The reporter expressed trepidation about the risks involved but proceeded with the experiment for the sake of journalism.
Meta's autonomous agent
Meta launched Muse in early September, positioning it as an agent that can handle emails, book travel, and make purchases on behalf of the user.
The tool operates in the background even when the user closes the app, raising concerns about how easily it can be deployed without constant oversight.
Meta uses Stripe Link to generate virtual cards for transactions, adding a layer of protection, but the safety mechanism is named "Sentinel," a reference to the genocidal robots in X-Men lore.
"We don't get this is a problem. We get, this is a problem, and here's why," said Kevin Turnbull, Global IT Director at Asala Energy.
The next step
As AI agents become more capable of performing complex financial tasks, regulators and financial institutions are racing to establish guardrails.
The central question remains whether current safety measures are sufficient to prevent widespread fraud in an increasingly automated digital landscape.
Related
