Canada's Big Banks Just Locked Out Your AI Agent. Their Own Are Coming.

Canada's banks spent a decade telling customers never to hand their password to anyone. This week they extended that rule to software. The Royal Bank of Canada told staff in an internal memo that "third-party personal AI agents accessing banking systems can create unacceptable customer risk," and became the first Canadian bank to block them from its online banking, The Globe and Mail reported on October 27.

RBC's lead lasted about a day. TD said it "has made the decision to block all third-party personal AI agents from direct access to banking information," according to Global News, and BMO, CIBC and Scotiabank announced their own limits on October 28, per Bloomberg Law and a Globe follow-up. National Bank says it is watching closely. In roughly 24 hours, the Big Five went from no policy to a shared wall.

What Set Off the Bank AI Agent Ban?

A consumer product that works the way agents are supposed to work. Meta launched its Muse assistant on September 8, and it can log in to sites on a user's behalf with the user's own credentials. RBC CEO Dave McKay said on Bloomberg TV that the bank was "considering blocking" Muse a week before the memo went out, The Logic reported. Amazon has already asked Meta to keep Muse off its store, according to Global News.

The US banks moved first. Morgan Stanley blocked third-party agents, JPMorgan and Bank of America are restricting them, and Citi says it "does not permit third-party AI agents to log into customer accounts or transact directly," according to a summary of reporting by Axios and Reuters. Those reports said some agents contacted bank call centers or appeared to act without their users' instructions. The same summary flags that this describes a risk rather than proof of widespread unauthorized transfers, and that a source close to Morgan Stanley said the bank had not seen rogue agents itself.

RBC's spokesperson Jeff Lanthier was frank about who is affected. "We understand that this may be disappointing for the small portion of our clients who use these tools," he told the Globe.

Is the Bank AI Agent Ban About Security or Control?

Both, and the security case is real. An agent that logs in as you is technically indistinguishable from you. Global News points out the uncomfortable consequence: when a customer shares their login and something goes wrong, the transactions may not count as "unauthorized," which is the line most fraud reimbursement depends on.

Ottawa agrees on that point. The office of Evan Solomon, Canada's Minister of AI and Digital Innovation, said Canadians "should be extremely cautious about sharing their banking passwords with anyone, whether that's an AI agent, an individual or a company."

BMO CEO Darryl White compared credential sharing to leaving sensitive information in a parking lot where "everybody walks by and reads it," according to the Globe. In a notice to clients reported by PYMNTS, RBC warned that agents "may store your passwords."

The control argument is harder to wave away. The same five banks blocking outside agents are building their own. All five are early issuer partners in Visa Agentic Ready, a program for testing payments started by AI agents, Visa Canada announced in May. TD's statement this week talked about "a future where trusted AI agents can create meaningful value for clients," and the operative word is trusted, as in approved by the bank.

The banks have also been inconsistent about credential sharing when it suited them. RBC, BMO and National Bank have all offered products that ask customers to enter logins for outside accounts, The Logic reported last year. Koho CEO Daniel Eberhard called the industry's security lobbying on open banking "preposterous" given how much worse screen scraping is.

Can Banks Keep AI Agents Out for Long?

The people who build this technology doubt it. "Agents aren't going anywhere, and firms on all sides of this transaction will need to adapt," Jeff McMillan, former head of firmwide AI at Morgan Stanley, told American Banker. His sharper warning is about competition. "The agents will go and open accounts for consumers at other firms that are not as restrictive."

Enforcement is also leaky. Transmit Security CEO Matt Thompson told American Banker that banks are "blocking agents that are self-identifying, but they all recognize that there are still agents getting through." American Banker notes the parallel with around 2015, when banks blocked screen scrapers like Plaid and Mint and then spent years negotiating API agreements with them instead.

The regulator is not going to settle it. OSFI Superintendent Peter Routledge told the Globe his approach to third-party AI risk is "principles-based," adding, "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."

How Does Open Banking Fit Into This?

It is the missing piece. About nine million Canadians already share financial data through screen scraping, "exposing them to security, liability, and privacy risks," the Department of Finance said in June when it pre-published the draft consumer-driven banking regulations. That framework would replace password sharing with permissioned access through APIs, with liability rules attached. It is the safe channel an AI agent would need, and it is not live yet.

So Canadians are left with an awkward gap. The banks have closed the unsafe door, and the safe one is still under construction. Customers are not exactly demanding agents yet. A June TD and Ipsos survey of 2,501 adults found 71% have more confidence in human intelligence than AI, while 50% said they would use AI for budgeting.

What Should Canadians Do Right Now?

Treat any app or assistant that asks for your banking password the way you would treat a stranger asking for it. If you already gave one access, change your password and turn on multifactor authentication. If you want help with budgeting, export statements or use tools that connect through your bank's approved channels.

The bigger question belongs to Ottawa. RBC's McKay put it this way, according to The Canadian Press: "If AI is going to move beyond the frontier and into the broader economy, trust will matter as much as capability." He is right, and the fastest way to build that trust is the open banking system the country keeps delaying. Until it arrives, the banks get to decide which agents count as trusted, and they will mostly pick their own.