Banking regulator won’t require banks to block AI agents from accessing client data

Canada’s banking regulator is ramping up oversight of artificial intelligence practices at the country’s largest lenders, but would stop short of barring financial institutions from allowing customers to use personal AI agents, such as Meta Platform Inc.’s Muse.

Royal Bank of Canada is considering blocking Muse from accessing the bank’s customer information, chief executive officer Dave McKay said in an interview last week with Bloomberg TV. Office of the Superintendent of Financial Institutions superintendent Peter Routledge said Tuesday he would ultimately leave the decision up to individual banks, so long as they are following the regulator’s guidelines.

“I don’t think we would get down to the level of you should do it this way, and so everyone should do it this way. I don’t think that’s not our highest and best use,” Mr. Routledge told reporters during the Tech Festival Canada in Toronto. “Where I stand right now on third-party risk and AI is we’re more principles-based.”

Customers can authorize AI agents, including Muse, to access their bank accounts, transaction history and credit card data to analyze spending patterns and make recommendations.

Mr. Routledge said customers could choose to switch financial services providers based on whether or not they want to use these types of services.

Earlier this month, Mr. Routledge said OSFI is planning to implement tighter guidelines on AI as threats from agents and Anthropic’s powerful Claude Mythos model accelerate risks among the country’s banks. In the spring, the regulator published guidelines for banks on agentic AI that ask whether the bank’s connection to a third party puts the institution or its customers at risk.

Global financial watchdog the Financial Stability Board has recommended financial institutions reinforce oversight of their AI systems, including restricting agents’ access to sensitive information.

Cohere catches Alibaba AI model trying to upload file to Beijing server

U.S.-based banks JPMorgan Chase & Co. and Charles Schwab Corp. recently said they are blocking or limiting outside AI agents’ access to their customer’s data.

Bank of Montreal CEO Darryl White said at an event last week that customers should not share sensitive banking data with AI agents.

“To the extent a board of directors and their CEO decide now we’ve got to mitigate this risk from agents that are perhaps issuing transaction instructions on behalf of a human being – maybe that board and that CEO decides that’s too risky,” Mr. Routledge said.

“Provided it’s done in a responsible way, that to me is indicative of sound risk management.”

Mr. Routledge’s seven-year term as the head of OSFI will come to a close in June, 2028, unless he is reappointed. In the lead-up to that deadline, he said his most significant tasks will be to develop a framework for managing AI risks and changing OSFI’s culture and its approach to risks.

The next superintendent’s term, which will run from 2028 to 2035, will have to focus on major forces that will transform financial services, Mr. Routledge said. That includes the digitalization of money and the arrival of AI.

He said the financial system in 2035 will look vastly different than the one that exists today.

The next superintendent “is going to have to figure out how to supervise in that period of immense change and deliver both resilience – which Canadians want from their financial system – and Canadians want more competition,” Mr. Routledge said.

“They want more services for lower costs, and they want a financial system that’s fuelling growth at a time of tremendous economic transition for the country.”