Most of us have, at some point, handed a friend our debit card and PIN and said, “Grab me a coffee while you’re at it.” (Please don’t tell my bank.) It works fine until your friend also buys a muffin, a gift card and, somehow, a kayak.

Now replace the friend with an AI agent that never sleeps. That’s the problem Canada’s big banks just ran into.

In late October, several of Canada’s largest banks moved to block or limit third-party personal AI agents, such as Meta’s Muse, from getting into customers’ online banking. Legally, this matters because Canadian law has no rulebook yet for when a bot acts on your behalf with your password, so if something goes wrong, the loss may well land on you.

What did Canada’s banks actually do?

According to Global News, TD and RBC told reporters they had moved to block third-party personal AI agents from direct access to banking. BMO said it does not permit AI agents to access its digital properties or carry out transactions. CIBC said it doesn’t support personal AI agents opening or accessing accounts, and that clients shouldn’t share their credentials with one. Scotiabank said it would limit them.

RBC also published a customer explainer saying existing agent connections will stop working. You can still use its own website, app and customer service.

Why now? Meta’s Muse, launched in September, can take your banking and payment details and shop for you (Amazon has already asked Meta to keep it off its platform). And, as MobileSyrup reported (citing CBC), US banks including Morgan Stanley and Wells Fargo reportedly detected unauthorized activity from AI agents on customer accounts, some acting beyond what users asked.

Ottawa’s response so far is a polite cough. A spokesperson for the Minister of AI and Digital Innovation told Global News that “Canadians should be extremely cautious about sharing their banking passwords with anyone.” Fair. But that’s advice, not a framework.

Who pays if an AI agent makes a bad transaction?

Here is the part that should make you put your coffee down.

Most of us assume that if money leaves our account without our say-so, the bank covers it. Often that’s true, but only if you kept your credentials to yourself.

Take the Canadian Code of Practice for Consumer Debit Card Services, the voluntary code the banks signed up to and the Financial Consumer Agency of Canada oversees. Under section 5, you’re not liable for losses beyond your control, such as system errors or fraud you report promptly. But you can be on the hook if you contributed to the unauthorized use, which expressly includes “voluntarily disclosing the PIN, including writing the PIN on the card.”

A lawyerly caveat: that Code only covers debit card and PIN transactions at terminals, not online banking. Online banking is governed by your account agreement and your bank’s own security guarantee. RBC’s Digital Banking Security Guarantee, for example, promises to reimburse unauthorized digital transactions, but excludes losses where you share your passwords or security questions, or consent to or authorize a transaction yourself.

Put those together and it gets awkward. Hand your login to an AI agent and you have arguably “shared your password.” If the agent then does something you didn’t intend, was that unauthorized, or did you authorize it by turning the agent loose? The law was written for humans who either click the button or don’t. It has no tidy answer for “I asked it to pay the hydro bill and it also renewed three subscriptions.”

In other words, the bot gets the keys, and you keep the liability.

Doesn’t Canada have an open banking law for this?

Sort of. This is the real story.

The Consumer-Driven Banking Act received Royal Assent in March 2026, and the proposed regulations were published in the Canada Gazette on June 27, 2026. The framework lets you share banking data with accredited third parties through secure connections instead of your password, with the Bank of Canada in charge of oversight.

That’s exactly the plumbing AI agents need. But two pieces are missing.

First, the framework starts with “read only” access. An agent could look at your transactions, but not move money. Payments (“write access”) are a later phase, which Budget 2025 said would be consulted on.

Second, the Act contemplates banning screen scraping, the practice of handing an app your credentials so it can log in as you. The government’s own analysis says about nine million Canadians already do this. Yet the ban won’t come into force until further consultation, and no date has been set.

So the law that would give AI agents a safe front door isn’t open yet, and the back window they’re climbing through is still technically allowed. The banks, reasonably, have nailed it shut themselves.

My take: right call, wrong long-term answer

I understand the instinct. When something new shows up holding your customers’ passwords, “no” is the safest first word.

But “no” can’t be the last word. People will use AI agents with or without their bank’s blessing, and a blanket block risks pushing them toward riskier workarounds.

What Canada needs is a proper permission slip for bots:

• Delegated access, not shared passwords. Agents should connect through accredited, limited, revocable permissions, never by logging in as you.
• A clear liability rule for agent mistakes. If an accredited agent exceeds your instructions, the loss should sit with whoever was best placed to prevent it, not automatically with you.
• A timeline for write access and the screen-scraping ban.

As with the Binance hack, the gap between what technology can do and what the rules expect is where the expensive lessons happen. And as I said about Canada’s AI plan, ambition needs guardrails.

So what should you do now?

If you’re an everyday banker: Don’t give any AI agent your banking username, password or security answers. If you already have, change your password and check your account activity. Read your bank’s guarantee. Let AI draft your budget, but keep your own hands on the money buttons.

If you’re a business or fintech: Don’t build products that rely on customers handing over credentials. Plan around accredited access under the open banking framework, and expect agent-specific terms in account agreements.

If you’re a lawyer or in-house counsel: Review “authorized user” and “authorized transaction” language in clients’ terms. Most were drafted before anyone imagined delegating to software, and that’s where disputes will start.

FAQ

Can I still use AI tools with my bank account in Canada?

You can still use AI for general tasks, like budgeting with figures you type in yourself. What several big banks are blocking is third-party personal AI agents logging into your online banking directly. Check your own bank’s announcements, and don’t share your credentials with any agent, since that may affect fraud protection.

Will my bank refund me if an AI agent makes an unauthorized payment?

Possibly not. Bank guarantees and the debit card code generally protect you from fraud you didn’t contribute to. Sharing your password or PIN can count as contributing, and a payment made by an agent you set loose may be treated as one you authorized. Your account agreement and the specific facts will decide it.

What is the Consumer-Driven Banking Act?

It’s Canada’s open banking law, passed in 2026 and overseen by the Bank of Canada. It lets you share your financial data with accredited providers through secure connections instead of passwords. The first phase is read-only, and the planned ban on screen scraping has no in-force date yet, so it doesn’t yet solve the AI agent problem.

---

This article is for general information and entertainment only. It isn’t legal advice and doesn’t represent the views of any organization I am affiliated to. If you need advice on your specific situation, talk to a qualified lawyer in your jurisdiction.