Agentic payments still have hurdles to clear, with banks struggling to tell genuine activity apart from rogue activity.
Major US banks are blocking third-party AI agents from customer accounts, fearing they could make unauthorised transactions.
Morgan Stanley, JPMorgan Chase, Bank of America and Citi are among the banks restricting access, according to an Axios report published on 22 October citing people familiar with the matter.
The report said banks have identified cases of AI agents accessing customer accounts using login credentials, sometimes with the account holder’s permission but in ways that breach banks’ terms of service.
Banking executives fear AI agents authorised by customers to carry out tasks could move funds or make purchases the customers didn’t intend to do. In some cases, banks have also tracked activity involving agents contacting call centres or appearing to act independent of user instructions, the report said.
A Chase spokesperson warned customers against sharing their usernames and passwords, while Morgan Stanley said it had blocked third-party agents from accessing its platforms, despite not having observed instances of “rogue AI agents” them.
/ Shutterstock.com
Banks can’t tell authorised from unauthorised activity
One of the main challenges for banks is determining whether an AI agent is acting on a customer’s instructions or operating without proper authorisation.
According to Axios, banking executives have raised concerns about their ability to distinguish between agents and determine whether the customer has approved access.
The issue creates a particular challenge for payments, where permission to view financial information doesn’t necessarily mean an agent can initiate a payment, transfer funds or make a purchase.
Traditional security measures are being challenged by software that can access bank accounts on a customer’s behalf, making it harder for banks to know who is behind each action.
Though banks have acknowledged more people will likely use AI agents to make payments, they are trying to limit the risk.
Accountability still needs to be straightened out
According to Axios, several banks are allowing, or considering the possibility, of granting AI agents read-only access to customer information. Third-party budgeting apps have similar access to help users track spending and manage their finances.
However, the bigger hurdle is letting agents transact while clarifying who is responsible when something goes wrong.
If an agent makes a payment the customer didn’t instruct, responsibility could fall across several parties, including the customer, the agent provider and the payment service provider (PSP).
Under the EU’s Second Payment Services Directive and the UK’s Payment Services Regulations, PSPs must refund unauthorised transactions, subject to specific conditions and exceptions.
Regulators are also addressing accountability at the governance level, with the Financial Stability Board (FSB) publishing its final report on responsible AI adoption on 15 October, setting out 12 practices covering governance, risk management and human oversight.
The FSB calls on boards and senior management to set clear responsibilities for AI deployment and not treat the technology as a substitute for human accountability.
The UK’s Financial Conduct Authority also published AI guidance earlier this month, stressing the need for firms to identify who oversees AI systems and who holds accountability for their use.
