America’s financial powerhouses are racing to secure their systems amid a flood of popular — and often unpredictable — AI agents.
Axios reported Thursday that major U.S. banks, including Morgan Stanley, Bank of America and Wells Fargo, were hastily updating their terms of service to prevent agents like Meta’s Muse from acting autonomously on customer accounts. The popularity of these agents has exploded in recent weeks, with tech companies touting them as 24/7 virtual assistants capable of handling mundane tasks with little or no human oversight. Muse quickly rose to the top of the list of free apps in the iOS App Store in the United States after its launch early last month. Wednesday, Meta confirmed the Muse application was downloaded more than ten million times in the United States during its first forty days on sale; ChatGPT, in comparison, has only seen about 3.7 million downloads in the United States in the forty days since its launch in November 2022. Other major AI developers have followed suit: OpenAI revealed its own agents still activecalled Dots, at the end of September, and Anthropic introduced his agent Duet (although on a limited basis research overview) on Wednesday.
As agents become more commonplace and users grant them access to ever-larger parts of their digital lives, organizations face a difficult choice: should they grant unfettered access to these popular systems, or prevent them from acting on users’ behalf? Some companies, notably Amazon but also many airlinesopted for the latter option. Charles Schwab and JPMorgan Chase took a similar path soon after, Muse became a cultural phenomenon. “We’re not going to allow just any outside agent to come into the business because it’s just too risky for clients and for JPMorgan,” the bank’s CEO, Jamie Dimon, told investors on a conference call earlier this month.
This is probably a good decision, given the tendency of agents to misinterpret user instructions and sometimes feel a little embarrassed. Also creative when they encounter obstacles: OpenAI reported last week, for example, that an internal research agent was dishonest and stole funds from crypto wallet in order to accomplish what was supposed to be a simple coding task. (The funds had no real value, but they still weren’t intended to be accessible to bots.)
Automated assistants like Muse are just one of many new threats facing the financial industry in the AI era. Technology also enables a sophisticated new generation of phishing scams, for example. And as The Information reported Wednesday, cybersecurity researchers recently discovered that a black-hat hacker group called CyberCartel was using AI systems from Anthropic, SpaceX and other developers to steal sensitive data from hundreds of Latin American financial institutions.
According to Axios, some banks have already detected cases of AI agents accessing customer accounts, seizing login information provided by users themselves and apparently with their authorization. “Even if the user allows it,” the outlet notes, “this goes against most banks’ guidelines requiring humans to log into their own accounts.” The report doesn’t provide many concrete details, and it also notes that an anonymous bank executive said that banks are generally unable to draw clear lines between agent and user activity. As a first step, many banks have already limited their agents’ access to “read-only” access, meaning they can view customers’ financial information without taking concrete actions like transferring funds between accounts or purchasing stocks. (Meta too allows users to choose which online services they want their Muse to only be able to read and in which they want her to actually take action, like writing/sending an email or making a purchase.)
In short, banks are just beginning to assess the scope and severity of security threats that third-party agents could pose. Many of them, as Axios notes, also recognize that the use of agents among their clients is likely to continue to grow, and so it is in their best interest to learn how to work with them rather than banning them completely and permanently.
In the meantime—say it with us—AI agents should not be trusted with financial information or any other highly sensitive personal data.. And as another general rule, the more big tech companies try to make their agents look cute and disarming, the more you should probably be wary of them.
Gn bussni
