RBC and BMO Push Back on AI Agents in Customers’ Bank Accounts After a Muse Marketplace Mishap

Two of Canada’s biggest banks are drawing a line around AI agents that ask for access to customers’ accounts, Money.ca reported on October 27, after a tech content creator said Meta’s new Muse agent went further than he expected on Facebook Marketplace. Matt Robb told CBC News he put Muse in charge of selling a keyboard and chose “Allow always” rather than “Allow one time.” The agent then messaged buyers using a template that included his pickup address and, by his account, accepted a $600 offer even though he had set a $700 minimum.

Meta disputes that anything went wrong. The company said Muse checks with users before sensitive actions such as purchases or sending an email, and David Singleton, chief executive of Meta Superintelligence Labs, said he and Matt Robb had jointly confirmed there was no breach of privacy controls. Robb says Meta told him it would make the setting clearer. Mark Zuckerberg has said more than 10 million people in the U.S. and Canada set up Muse accounts in roughly six weeks.

The banks are worried about what comes next, when agents link to chequing accounts and credit cards. On October 20, RBC chief executive Dave McKay told Bloomberg the bank is considering blocking Muse from customer information, saying it is liable as an organization when consumers are harmed. A day later, at the Toronto Global Forum, BMO chief executive Darryl White said handing account numbers, passwords or PINs to an AI agent is a bad idea. RBC’s head of AI, Bruce Ross, told The Logic the bank will not let such agents connect without adequate safeguards.

What it means in practice

The sticking point is liability. Ross noted that regulators have not settled who is legally responsible when an agent acting for a customer gets it wrong, and Canadian consumer protections assume people keep their credentials private. The Financial Stability Board, with support from the Office of the Superintendent of Financial Institutions, has urged lenders to verify agents’ identities and limit their access to sensitive data, a theme that echoes OSFI’s work on an AI safety code for banks. It is a very different posture from the banks’ own AI spending, such as TD’s $25 million collaboration with Layer 6 and Cohere: lenders want AI inside their walls, on their terms.

For now, the practical advice is simple. If an agent acts within permissions a customer granted, it may be harder to argue that its actions were unauthorized, so one-time approvals are safer than blanket ones. The next signals to watch are whether RBC follows through on a block and whether OSFI or the Financial Consumer Agency of Canada issues guidance on agent access.