Major U.S. banks have reportedly detected a series of cases where unverified artificial intelligence (AI) agents were operating in customer accounts and have begun taking steps to block them, according to the U.S. online media outlet Axios.
Citing sources, Axios reported on the 22nd local time that some AI agents exhibited behaviors seemingly unrelated to customer instructions or even directly contacted bank customer service centers.
In most cases, actual users—the customers themselves—had handed over their login credentials to AI agents, allowing them to access their accounts.
However, even with the user's consent, this violates the regulations of most banks that require customers to access their accounts directly.
Industry sources noted that banks are particularly concerned that AI agents granted payment and money transfer privileges could take actions affecting customer assets.
The biggest issue is that banks struggle to properly distinguish between humans and bots, as well as between AI agents whose activities were approved by customers and those that were not.
In response, Morgan Stanley has started blocking AI agents from accessing its platform.
JPMorgan Chase, Bank of America, and Citi are also restricting access.
Banks are also rushing to revise their terms of service.
Most banks are moving toward allowing "read-only" access, where AI agents can view financial information without executing transactions.
While acknowledging that a compromise is needed given customer demand for AI agent usage, the industry currently judges that the risks of allowing them too freely are greater, Axios reported.
※
Copying, redistribution, and unauthorized use in AI training are strictly prohibited.