Two years ago, generative AI in a brokerage meant a chatbot that could draft an email and occasionally hallucinate a coverage that didn’t exist. Today, the same type of tool can work through a multi-step renewal file, cross-reference a schedule of values, flag the discrepancy, and draft the client email, largely unsupervised, for hours at a stretch.
Regulators are working hard to keep pace but it’s an uphill battle. Rulemaking runs on a timeline built for a world where the technology didn’t change between the draft and the final version. AI doesn’t wait for anyone’s comment period. For a brokerage owner, that gap between what AI can do and what any rulebook says about it is already the space you’re operating in, whether you’ve deployed AI or not.
1. The Capability Curve: What Actually Changed
The real shift wasn’t intelligence. It was autonomy.
Two years ago, these tools needed a prompt for every step, like a capable intern who still needed instructions for opening a door. The current generation of models from major providers, including OpenAI, Google, and Anthropic, can now sustain multi-hour, multi-step work with minimal supervision, reason through novel problems rather than pattern-match, and process voice, text, and documents in the same conversation. The distance between “a tool that answers questions” and “an agent that completes a workflow” has closed fast enough that most compliance frameworks haven’t caught up to what changed.
Canadian insurtechs built specifically for the broker channel have kept pace.
- Quandri’s Renewal Intelligence Platform now automates policy checking and requoting.
- Cemiar allows data integration and API automation between brokers and carriers as well as producing capable AI agents.
- Koïos Intelligence’s Olivo platform runs multi-agent conversational AI across the full policy lifecycle, from quoting to renewal.
- Trufla Technology recently launched what it’s calling the industry’s first end-to-end broker platform with embedded AI.
None of these are experimental anymore. They’re production tools inside Canadian brokerages today.
2. Where Canadian Regulation Actually Stands
Here’s where each regulator has landed, and it’s a good start.
RIBO published its “Responsible AI Use Among RIBO Licensees” guidance in May 2025, following a study it commissioned from the Behavioural Insights Team. It’s principles-based rather than prescriptive, a reasonable choice given how fast this technology keeps changing. It reminds licensees that AI can’t replace professional judgment and that the existing Code of Conduct and Fair Treatment of Customers obligations apply whether or not AI is involved.
Quebec’s AMF released a draft AI guideline in June 2025 covering governance, lifecycle risk management, and fair treatment. It’s more structured than RIBO’s approach, though still in draft form, which for a technology this young is arguably the more cautious way to get it right.
OSFI has gone furthest on paper. Its finalized Guideline E-23 on Model Risk Management, published September 2025, explicitly captures AI and machine learning under a broad definition of “Model” and applies to all federally regulated insurers, life and P&C alike. It doesn’t take effect until May 1, 2027, an 18-month runway that gives institutions time to actually build the compliance muscle instead of bolting something on overnight. In July 2026, OSFI followed up with a bulletin specifically on generative and agentic AI, flagging that autonomous systems now operate with little human supervision and layer new risk onto existing technology, cyber, and third-party risk frameworks.
FSRA hasn’t issued AI-specific guidance for Ontario insurers, and given how quickly “AI-specific” tends to become outdated, that’s arguably the more defensible position. Its June 2026 Operational Risk and Resilience Guidance covers cyber, third-party, and climate risk broadly, but doesn’t address AI directly, a sign FSRA is still watching this space, rather than directly regulating it.
3. The Gap You’re Actually Standing In
Add it up: RIBO’s guidance is non-binding. The AMF’s is still a draft. OSFI’s binding rule doesn’t land until 2027. FSRA has no AI-specific rule at all. And at the federal level, there is no AI statute in force. The Artificial Intelligence and Data Act (AIDA), Canada’s attempt at comprehensive AI legislation, died on the order paper when Parliament was prorogued in January 2025 and has not been reintroduced. Ottawa’s June 2026 national AI strategy, “AI for All,” chose not to revive it, opting instead for a slower, multi-bill approach built around sovereign compute investment and adoption incentives rather than binding rules. None of that is controversial. It’s what happens when a technology’s release cycle is measured in months and a legislative cycle is measured in years.
However, none of that changes your liability. RIBO’s Fair Treatment of Customers standard, the Code of Conduct, and E&O exposure all apply today regardless of whether a specific AI rule exists to point to. There’s still a standard. It’s just being enforced after the fact, through complaints and E&O claims, instead of in advance, through a compliance checklist you could follow.
4. What This Means Before the Rules Arrive
Waiting for OSFI’s 2027 deadline, or for RIBO to formalize something more binding, is a bet that nothing goes wrong in the meantime, not a strategy. Brokerage owners who are ahead of this aren’t waiting for either rule to become mandatory; they’re voluntarily adopting RIBO’s May 2025 guidance and OSFI’s E-23 principles as their working standard now. In practice, that means documenting which AI tools are in use, requiring vendors like Quandri, Cemiar, Koïos, Trufla and others to answer direct questions about data handling and model behaviour, and keeping the human-in-the-loop discipline this series has covered before, not because a regulator has required it yet, but because the professional standard already does.
Summary: The Rules Are Still Catching Up
The real risk is that “slow” gets misread as “permissive”. Regulators aren’t dragging their feet, they’re chasing a technology that moves faster than any rulemaking cycle was built to handle. OSFI’s 2027 deadline and RIBO’s non-binding guidance are the floor you should already be building toward, not a grace period. A brokerage that waits for the rulebook to arrive is choosing to find out where the lines are through a client complaint or an E&O claim, rather than through a policy it wrote itself.
