According to the majority of internal accounts, the last three years of AI deployment in Canadian banking have seemed like a clear success story inside the glass towers of Bay Street. Applications for mortgages that used to take days to process now clear in a matter of hours. Algorithms manage credit evaluations that need to be reviewed by an underwriter before a human has finished their morning coffee.
AI implementation has saved 1.2 million working hours, according to CIBC alone. The cost savings and productivity figures are real, and the push to automate has only increased due to the aggressive entry of non-bank lenders into the same sector. A more straightforward question is currently causing unease in a few of boardrooms: were some of those quicker approvals made for the wrong reasons?
| Category | Detail |
|---|---|
| Primary Regulator | Office of the Superintendent of Financial Institutions (OSFI) — identified mortgage risk as Canada’s top financial system threat in 2026 |
| AI Productivity Gains Claimed | CIBC reported saving 1.2 million working hours through AI tools; TD Bank reduced mortgage approval processing times using automated underwriting |
| Mortgage Renewal Risk Window | Approximately 52% of total Canadian mortgages are expected to renew between January 2026 and end of 2027 — many borrowers face significantly higher payments at renewal |
| Non-Bank Lending Concern | OSFI flagged competition from private, non-bank lenders as pushing regulated banks toward riskier lending practices to retain market share |
| Consumer Stress Indicator | Bank of Canada data shows rising delinquency rates in auto loans, particularly among younger borrowers — an early indicator of broader credit stress |
| AI Limitation Identified | Machine learning credit models trained on historical transaction data may fail to anticipate unprecedented economic shocks — specifically rate-driven payment stress not well represented in training data |
| Bank Investment Outlook | RBC and TD both project substantial enterprise value from AI tools by 2027–2028, indicating continued investment despite rising default concerns |
| Further Reference | Canadian banking risk analysis at Bank of Canada Financial System |
Mortgage risk is the biggest threat to Canada’s financial system as of 2026, according to the Office of the Superintendent of Financial Institutions, which oversees the nation’s main banks. The timing is precise. About 52% of all outstanding Canadian mortgages are anticipated to be renewed between January 2026 and the end of 2027. This wave of refinancing would compel hundreds of thousands of borrowers to make payments that are far higher than what they originally agreed to.
The amount of household debt that accumulated over the years of low interest rates is no longer theoretical. The AI models that assisted in approving many of those initial loans were not created with this particular stress situation in mind, and it is making its way to kitchen tables in planned monthly increments.
This is the main technical issue with using machine learning to credit underwriting during a time when interest rates are changing quickly. These models are trained using historical data collected over an extended period of low and comparatively constant interest rates, including transaction records, payment histories, and income trends. They are truly optimized for a world that has vanished.
The model’s confidence intervals may become subtly incorrect when the economy drastically changes, resulting in approvals that appear statistically sound when compared to the training data but perform poorly when the data never predicted. This trend may account for some of the increased default outlook: loans that would have been handled more cautiously by a human underwriter with a more comprehensive understanding of macro risk yet passed computer examination in 2022 or 2023.
With an emphasis on how credit risk is evaluated, tracked, and alerted when loans start to perform poorly, OSFI has begun having direct discussions with banks regarding their AI underwriting systems. Concerns have also been voiced by the regulator over the pressure banks are under from private, non-bank lenders that operate with fewer regulatory restrictions.
This competitive dynamic offers incentives for banks to relax standards, regardless of whether an algorithm or a human approves them. Although the combination of market pressure to compete and AI-speed approvals is not intrinsically harmful, it does necessitate the kind of strong risk management infrastructure that is typically developed and implemented even more slowly than the technology it is meant to regulate.
Observing Canadian banks negotiate this at the same time gives the impression that a key phase of the industry’s growth will be defined by the conflict between the efficiency case for AI lending and the regulatory cautious case. By 2027 and 2028, both RBC and TD anticipate significant returns from AI investment.
AI-approved portfolios are embedded in a financial system that is already under early-stage strain, according to the Bank of Canada’s consumer stress data, which shows an increase in vehicle loan delinquencies, especially among younger borrowers. It is really unclear if the default curve will rise gradually enough to be managed or if it will spike in a way that necessitates a more comprehensive recalibration of how AI-driven credit decisions are reviewed and challenged.
The algorithms continue to operate. The approvals continue to be processed more quickly than any human staff could. The tide of regeneration is intensifying. It is difficult to ignore the fact that the banks that developed these systems the quickest are now devoting a significant amount of effort to figuring out exactly what they created.
