Canada’s Banks Need an AI Outage Plan, Not Just an AI Policy

Matthew Leo · Published August 31, 2026 · Canada Tech

Bank operations staff responding to a system outage in a Toronto control room at dawn

Canadian financial institutions have spent years writing policies for how employees may use artificial intelligence. The next useful document is less glamorous: a plan for what happens when a critical AI or cloud provider is compromised, unavailable or behaving unpredictably.

That need became harder to dismiss on August 31. Financial Stability Board chair Andrew Bailey warned G20 finance ministers and central bank governors that frontier AI’s effect on cyber risk is the most immediate AI concern for the financial system. The FSB said advanced models could alter the speed, scale and economics of attacks and pointed to financial institutions’ dependence on a small group of technology providers as a systemic vulnerability.

Canada has three institutions represented on the FSB: the Department of Finance, the Bank of Canada and the Office of the Superintendent of Financial Institutions. This is not an abstract foreign-regulator debate.

The concentration problem is already familiar

A bank can diversify its branches, staff and internal applications while still depending on the same handful of cloud, identity and cybersecurity platforms as its competitors. An incident at one provider can therefore become an industry problem without spreading from bank to bank in the traditional sense.

AI adds another layer. Models and agents may sit inside fraud detection, customer service, software development, security operations and internal knowledge systems. The more authority they receive, the more important it becomes to know what happens when the model, its credentials or the infrastructure behind it cannot be trusted.

An outage plan should start with dependencies

Every critical AI-enabled workflow should have an owner and a documented dependency chain: model provider, cloud region, identity service, data source, API gateway and any external tool the agent can call.

Then classify what happens when each dependency disappears. Can staff fall back to a manual process? Does the service fail closed? Can an AI agent’s credentials be revoked without breaking unrelated systems? How long can the business operate that way?

Recovery should be rehearsed without the vendor

Canada’s participation in the G7 2026 cross-border cyber exercise is useful context. The exercise focused on response, recovery and crisis communication after a major financial-sector cyber incident. That is the right direction because a resilience plan that has never been exercised is still mostly a theory.

Organizations should run at least one scenario where the compromised party is not the bank itself but a shared technology supplier. The exercise should assume normal support channels are overloaded and that the provider’s status page is not enough to make a business decision.

Five controls worth testing

The Canadian Cyber Centre’s frontier-AI guidance already tells organizations to assign ownership, maintain human involvement for high-impact decisions, limit what automated systems can do independently, use strong authentication for non-human identities and preserve the ability to intervene or disable automated actions.

Those recommendations become much more useful when converted from governance language into an outage drill.

AI policy answers who may use a tool and under what rules. An AI outage plan answers the nastier question: what can the institution still do safely when the tool it depends on is suddenly the thing it cannot trust?

Sources: Financial Stability Board, August 31, 2026; Bank of Canada: G7 2026 cyber exercise; Canadian Centre for Cyber Security: Frontier artificial intelligence.

Tags: Canada, banking, AI, cybersecurity

Read on Mapletechie