US banks scramble to restrict AI agent activity
A new report claims that major U.S. banks have now detected AI agents accessing customer accounts, and that some are even attempting to make transactions. Axios reports that it's hard to differentiate between an actual human and an unauthorized bot, which is why some institutions, including Morgan Stanley, Bank of America, JPMorgan, and Wells Fargo, are blocking the technology and changing their terms of service to account for this.
While an AI agent or AI assistant can access an account with a user's permission and through the use of their login credentials, this often goes against bank policy that requires users to maintain control of their account and not share it with anyone else. One of the fears that executives have is when an AI agent with login access takes unauthorized steps, like spending money or moving funds around without the explicit instructions of the account owner. There have even been situations where an AI agent called the bank directly and took actions independent of direct user command. Even though it wasn't explicitly told to do that transaction, it considered it a crucial step to accomplish the task assigned to it.
One hypothetical example of this is when a user tells an AI agent to purchase an item but gives it a credit card with no available limit. The AI agent could then have accessed the user's bank account to move some funds from whatever account it sees that has available funds to then effect the transaction, even if the user did not explicitly tell them that they could use the money in that said account. This is just one example where an AI agent's transaction could go wrong, which is why several U.S. banks are now restricting third-party AI agents from accessing their platforms. Nevertheless, they still recognize that their customers might want to use tools like AI agents to monitor their finances. This is why some institutions have given or are moving towards read-only access for AI agents, which is something that they've been doing for years already.
The move comes against a background of burgeoning AI agent use, coupled with the risk that agents can and often do act beyond their remit, making mistakes or performing tasks without instruction, often with disastrous results. An unsupervised AI agent recently deleted a TV show's entire edit weeks before broadcast, forcing the program off air. AI agents are notorious for deleting databases and directories, making it a small leap to imagine they cause some havoc with unfettered access to a bank account.
Financial institutions have deep experience when it comes to employing new technology while ensuring that their critical infrastructure remains safe - something that AI labs lack, according to a former OpenAI safety employee. Despite that, banks are still concerned about the speed at which AI technology is developing, fearing that the technology's capabilities might outrun their ability to control it, which is why they're preparing for worst-case scenarios and even planning for day after" scenarios after a catastrophic AI event.