Finnish banks bar AI agents from customer accounts as payment rules lag

OP and Nordea require customers to authenticate themselves and prohibit AI agents from accessing accounts or making payments, while assessing whether tightly controlled agent-based services could be introduced.

Banks in Finland are drawing a clear line around the use of artificial intelligence agents in personal finance, even as the technology gathers pace elsewhere. According to Iltalehti, OP says customers must log in themselves and must not hand their banking credentials to another person or to an AI agent. The bank’s technology and development director, Kasimir Hirn, told the outlet that breaches of the terms can lead to account access being closed.

That position means an AI agent cannot, under OP’s current rules, enter online banking on a customer’s behalf to review transactions or make payments. OP says it is monitoring the development of the technology and is building tools to detect and block programmatic use of credentials. Hirn also said the bank is continuing to assess whether it could eventually support controlled access for corporate customers through AI interfaces, but added that no timetable or implementation decision has been announced.

Nordea has taken a similar approach. The bank told Iltalehti that, for now, customers must authenticate themselves and that AI agents cannot use its banking services or make payments on their behalf. At the same time, Nordea said it sees significant potential in AI-enabled services, while acknowledging that security, customer data and liability questions still need to be resolved. The bank said its assessment of agent-based AI is being guided by a responsible approach, with safety and trust remaining central to any future development.

The cautious stance is not limited to Finland. Reuters, citing Axios, reported that major US banks have begun restricting third-party AI agents from accessing customer accounts and carrying out transactions. The report said banks have seen cases in which agents contacted customer service or appeared to act beyond users’ instructions, and that it can be difficult to tell when an account activity is truly being directed by the customer. That uncertainty is one reason financial institutions are moving carefully.

At the same time, the industry is not rejecting the concept altogether. Nordea has separately shown a live pilot in Finland in which an AI agent completed an online purchase and payment using Mastercard Agent Pay, with verification and token-based controls meant to keep the customer in charge. The International Monetary Fund has also warned that agentic AI in payments raises unresolved issues around privacy, cybersecurity, authentication and legal responsibility, especially because payment systems were built for people who explicitly authorise each transaction.

The broader debate has also been sharpened by Elon Musk, who has encouraged users to let xAI’s Grok access bank accounts and has suggested uploading credit card images so the system can buy things on a user’s behalf. A community note attached to one of his posts warned against sharing card details and pointed to xAI’s own cautions about entering personal information into prompts. The contrast underlines the central dilemma for banks: AI agents may offer convenience, but the security model for money management remains unsettled.