Canada stopped hedging. OSFI’s AI Bulletin plus the finalized E-23 Guideline pull generative and agentic AI squarely inside the model risk regime for every federally regulated financial institution, effective May 2027. The AGILE framework replaces EDGE, and E-23 now reaches any system that “processes input data to generate results”, third-party vendor models and fourth-party software included. This is the Canadian answer to the same agentic AI governance vacuum the FSB and the US Conference of State Bank Supervisors are closing simultaneously. Three regulatory tracks, one direction: technology-neutral, agentic-aware.
The charter wave tightened in the same fortnight. Revolut won a conditional OCC de novo charter, the second foreign fintech this year after Nubank, while Chime moved to acquire Stride Bank outright, ending the rent-a-charter era for scaled neobanks. Closer to home, Questrade secured Canada’s first fintech Schedule I bank license and launched an explicitly “agentic-first” posture on a Temenos cloud core. Charters are converging with proprietary intelligence: Revolut pairs its license with PRAGMA, a transformer foundation model trained on 40 billion events via NVIDIA; Questrade ships Flows; Malaysia’s Ryt Bank runs an in-house LLM as the primary customer interface under regulator approval.
National Bank of Canada, meantime, closed key approvals to acquire Laurentian Bank’s retail and SME portfolios: $7.5 billion in deposits, asset-light, no branches transferring. It is a deposit-scale play in direct response to the charter competition.
Why it matters: the defensible stack is clarifying into three layers, the charter, the proprietary intelligence layer, and the control architecture. OSFI, FSB, and CSBS agree the control layer is where trust gets built, and Deloitte’s finding that 83 percent of customers are anxious about agents acting without approval makes structural controls, consent gates, immutable audit, kill-switches, the moat, not the chatbot. The compliance clock is at month twelve of twenty. Banks still treating AI as back-office augmentation are betting against the direction every regulator and every newly chartered competitor is already moving.