Lenders including Morgan Stanley, JPMorgan Chase and Citi are restricting third-party AI agents that log in with customers’ credentials, struggling to tell human users from software and weighing read-only access as a compromise.
Â
Some of America’s largest banks are cracking down on third-party artificial intelligence agents that sign in to customer accounts, as they grapple with a security problem they have not faced before, according to a report by Axios. Software that carries out tasks on a customer’s behalf is becoming popular, but lenders say they often cannot tell whether a human or a machine is behind the activity on an account.
People familiar with the matter told Axios that Morgan Stanley, Bank of America, JPMorgan Chase, Wells Fargo and Citi are all working to contain the risks. Morgan Stanley has begun blocking outside agents from its platforms, while JPMorgan, Bank of America and Citi are also placing limits. Banks are also revising their terms of service to address the issue.
Why banks are alarmed
In many cases, agents get in using usernames and passwords handed over by the account holders themselves. Even when a customer agrees to this, it breaks the rules of most banks, which require people to log in personally. A Chase spokesperson urged customers not to share their credentials with anyone.
The worry goes beyond viewing balances. Executives fear that an agent permitted to make purchases or move money could take steps with direct financial consequences. Banks have also reportedly seen agents phone customer service centres or behave as though acting independently of their users’ instructions.
Distinguishing between agents is another obstacle. According to Axios, some banks struggle to establish whether a particular agent was authorised by a customer or gained access without permission. Even where a customer initially approved an agent, it can be hard to judge whether later actions stay within that permission. The pace of AI development compounds the difficulty, exposing weaknesses faster than institutions can respond.
Not all banks describe the threat in the same terms. A Morgan Stanley spokesperson said the firm had not seen “rogue AI agents accessing our platforms”.
Read-only access as a middle ground
Banks accept that customers will increasingly want AI help with financial chores, so several have approved, or are moving towards, read-only access. Such an arrangement would let agents view financial data without starting transactions or otherwise acting on accounts. Banks already allow comparable access for budgeting tools such as Quicken.
Industry and banking executives told Axios that lenders recognise the need for a balance that lets customers benefit without exposing accounts to unacceptable risk. For now, though, executives believe the dangers of unrestricted access outweigh the benefits.
Banks are also preparing for the possibility that their defences fail. According to sources cited by Axios, some are planning for the immediate aftermath of an incident, including how to quickly secure accounts or restore customer access if an agent causes disruption. Banks have deep financial resources and experience protecting critical infrastructure, but one source said money alone may not solve the problem given how fast the technology is moving.
The challenge for lenders, as agents grow more capable, is to permit legitimate uses while preventing automated systems from compromising accounts or acting in ways customers never intended.
