OSFI Weighs an AI Safety Code for Banks as Frontier-Model Risks Grow
Canada’s banking regulator is moving away from a hands-off approach to artificial intelligence. Speaking at the Global Risk Institute Summit in Toronto on Wednesday, Superintendent Peter Routledge said the Office of the Superintendent of Financial Institutions (OSFI) is considering an AI “safety code” that would set minimum standards for how federally regulated banks use the technology, The Logic reported.
Routledge tied the shift to the speed of technical progress. He pointed to Mythos, Anthropic’s latest frontier model, and to other powerful systems that could be turned to cyberattacks or illicit activity, as well as to autonomous AI agents that operate with little human oversight. “That’s frightening, and that is a risk that is front of center for us,” he told reporters.
The work is at an early stage. Routledge said there is no draft yet and no formally approved timeline, although a consultation could open as early as spring. He said the code would go further than OSFI’s earlier AI advisory reports and would likely build on the regulator’s work with the Global Risk Institute’s Financial Industry Forum on AI and with the Financial Stability Board.
What it means in practice
OSFI is aiming for a floor, not a ceiling. Routledge said Canadian banks are generally prepared for AI-related risks, but some are further ahead than others, so the regulator wants a “minimum level of safety or resilience” that stops weaknesses at one institution from spreading to its counterparties. He added that OSFI has enough staff overall but will prioritize hiring specialists in emerging risks, particularly AI and digital technology, at a time when federal agencies face pressure to cap headcount.
A code for the financial sector would add one more layer to a broader federal push, from Ottawa’s warnings about rogue AI risks to the new National Council on Artificial Intelligence, which is meant to guide how the AI for All strategy is put into practice. Researchers have also called for tiered AI transparency rules in Canada.
Routledge spoke in the same week that the Financial Action Task Force downgraded its rating of Canada’s anti-money-laundering supervision of financial institutions, a finding he said he did not dispute. What to watch: whether OSFI publishes a consultation draft in the spring, which uses of AI it treats as highest risk, and how the code lines up with international work at the Financial Stability Board.