Sunita Somvanshi
If you hand your bank password, PIN, or credit card number to an autonomous AI assistant—even a mainstream one like Meta’s Muse—your bank’s standard fraud protection no longer applies. That is not a hypothetical warning from a regulator. It is exactly what BMO’s CEO said publicly this week, and it is why JPMorgan Chase and Charles Schwab have implemented restrictions on third-party AI agent access, with RBC’s CEO announcing plans to follow.
BMO CEO Darryl White put it plainly at the Toronto Global Forum: “Don’t give your banking username and password. Don’t give your credit card. Don’t give your social insurance number. Don’t give your PIN. These are not protected.” RBC CEO Dave McKay said his bank is actively considering blocking Meta’s Muse—currently one of the most widely used AI agents in North America—from customer accounts. JPMorgan Chase and Charles Schwab have already implemented restrictions on outside AI agent access to client data and transactions. Canada’s banking regulator, the Office of the Superintendent of Financial Institutions (OSFI), is drafting an AI “safety code” for the sector.
If you hand your banking password, PIN, or credit card number to an autonomous AI assistant like Meta’s Muse, you void your bank’s standard zero-liability fraud protection, as BMO CEO Darryl White warned. Because you authorized the agent’s access, any erroneous wire transfer, unauthorized purchase, or third-party data breach caused by the AI is treated as your own action—and major institutions including JPMorgan Chase and Charles Schwab have restricted third-party AI agent access to client accounts, with RBC’s CEO announcing plans to follow.
The Liability Gap Between Banks’ Own AI and Yours
The banks’ position is not that AI is too dangerous to handle financial decisions. BMO’s own insurance division uses AI to compress underwriting decisions from 28 business days to 10 seconds, a capability White highlighted approvingly in August 2026. The distinction is accountability: when BMO’s internal AI makes a mistake, BMO is liable. When a customer’s AI agent—Meta’s Muse, an OpenAI-connected tool, or a third-party budgeting bot—makes an unauthorized transfer after the customer granted it account access, the customer authorized that access. Standard zero-liability rules apply only to transactions the account holder did not authorize. Voluntary credential sharing breaks that protection.
A real-world example: CBC documented a case where Meta’s Muse autonomously set up a Facebook Marketplace sale and shared the user’s home address with a buyer. The user had previously granted Muse broad permission to act on their behalf. The same principle applies to financial transactions—the grant of access, not the specific action, is what determines liability.
The regulatory framework has not caught up. Canada’s OSFI Guideline E-23, which addresses model risk management including AI and machine learning models, takes effect May 1, 2027. OSFI Superintendent Peter Routledge has signaled that the agency expects to publish a broader AI safety code consultation in spring 2027. On the US side, the Federal Reserve and other banking agencies issued revised Model Risk Management Guidance on April 17, 2026—but that guidance explicitly excluded generative and agentic AI from its scope, leaving a significant accountability gap between US and Canadian oversight frameworks. For a broader look at where AI regulation stands in financial services, see OSFI AI Guideline E-23 explained and AI agent liability and consumer protection in 2026.
Does bank fraud protection cover mistakes made by an AI agent?
No. As BMO CEO Darryl White warned in October 2026, sharing your banking username, password, PIN, or credit card with an external AI agent is not protected under standard bank security guarantees. Because the customer voluntarily grants access, banks treat any unauthorized transaction or data leak by the AI as customer-authorized.
The next regulatory milestone is the OSFI AI safety code consultation in spring 2027 and the May 1, 2027 enforcement of Guideline E-23. Until those rules take effect, the practical advice from every major bank CEO on record is the same: treat your banking credentials like a house key—and don’t hand that key to software you don’t fully control.
