There is no AI rulebook for wealth management in Canada or the US. What there is, instead, is a steady body of staff notices, regulatory notices, exam priorities and enforcement actions that all make the same basic point: the rules a firm already follows still apply when the work is done by an AI system.
That sounds reassuring, but it cuts both ways. A firm can't wait for an AI-specific rule before putting controls in place, because regulators are already examining AI use against existing obligations. And a firm that uses AI to research, summarize or monitor products is still fully responsible for its understanding of those products.
This article sets out what regulators in both countries have actually said, in their own words, and what it means for firms using AI in product due diligence and monitoring. It covers the Canadian Securities Administrators, CIRO and OSFI in Canada; FINRA and the SEC in the US; the themes they share; and a practical set of responsibilities and an example written process for AI in a Know-Your-Product (KYP) program.
It reflects publications available as of the date above. This is an active area, and firms should check for newer guidance.
Canada's securities regulators have published the most detailed guidance of the two countries, and it speaks directly to registered dealers and advisers.
In December 2024 the CSA published Staff Notice and Consultation 11-348, Applicability of Canadian Securities Laws and the use of Artificial Intelligence Systems in Capital Markets.[1] It is staff guidance on how existing law applies, not a new rule. Its starting point:
It adds an important qualification: "Depending on the technology, different actions may be necessary to meet the same requirements under securities law."[1] In other words, the obligation stays the same, but the controls needed to meet it change with the tool.
| Theme | What CSA Staff Said |
|---|---|
| Governance | "Governance and risk management practices should be paramount" for regulated market participants deploying AI systems. |
| Fit for purpose | Policies should cover planning, design, verification and validation, "including being satisfied that the AI system is fit for purpose and that robust testing prior to deployment has taken place." |
| Human in the loop | Policies can include "having a 'human-in-the-loop', where humans can effectively monitor the input and/or output of an AI system." |
| Ongoing testing | Staff "emphasize the importance of regularly testing the AI system and the results of its use before and after its adoption." |
| Explainability | Firms should balance advanced capabilities against explainability; staff favour "the highest degree of explainability that is feasible" for the type of system. |
| Record keeping | AI systems "should provide an appropriate degree of explainability so that registered firms are able to meet applicable record keeping requirements." |
| AI washing | Inaccurate, misleading or embellished claims about AI use "may be misleading to the public or constitute a misrepresentation." |
For dealers and advisers, the notice distinguishes between how much decision-making is handed to the system:
For product research and due diligence, the decision-support category is the one that matters most. An AI system can read, extract and summarize, but the firm must verify what it produces and make its own assessment.
A year after 11-348, the CSA and CIRO published Staff Notice 31-368, the results of their review of 105 firms' KYP and related practices.[2] It doesn't address AI as a separate topic, but it speaks to automated systems in two places that apply directly to AI tools used in product review:
The same review found that "annual monitoring alone was not found to be sufficient" for products on the shelf.[2] This is one reason firms are turning to automated monitoring. It also means the automation itself becomes part of what the firm has to supervise and evidence.
CIRO. CIRO's 2026 Compliance Report identifies AI as a focus area for dealers. It expects firms to understand and document where they use AI across the business, to have policies and controls for the risks it creates, and to consider whether a new use of AI or automation is a material business change that requires advance notice to CIRO. CIRO's compliance teams will review operational controls to confirm AI tools are working as designed.[3]
OSFI. Many Canadian dealers and advisers sit inside a bank or insurer. For those groups, OSFI's Guideline E-23 on model risk management, published in September 2025 and effective May 1, 2027, applies to federally regulated financial institutions.[4] It defines a model to include AI and machine learning methods, covers models and data sourced from third-party vendors, and requires a comprehensive, current inventory of models whose risk is not negligible.[4] An AI tool used to screen or monitor products inside a bank-owned firm is likely to fall within the group's model risk framework.
US regulators have taken the same technology-neutral line, delivered mostly through FINRA guidance, SEC exam priorities and enforcement.
In June 2024, FINRA issued Regulatory Notice 24-09, reminding member firms of their obligations when using generative AI.[5]
Two points in the notice matter for product due diligence:
FINRA's 2026 Annual Regulatory Oversight Report, published in December 2025, returns to the topic.[6] It observes that the most common use of generative AI among member firms is summarization and information extraction, which is exactly the work AI does in reading offering documents and filings. Among the practices it describes are formal review and approval of AI use cases, a supervision, governance or model risk management framework, "robust testing" of capabilities and limitations, and "ongoing monitoring of prompts, responses and outputs," including "validation and human-in-the-loop review."[6]
The report also flags a newer risk: AI agents that act autonomously, without a person validating each step. FINRA points to the risk of agents acting beyond their intended scope, and to the difficulty of auditing what they did.[6]
| Date | Action | What It Means |
|---|---|---|
| March 2024 | Settled charges against two investment advisers, Delphia and Global Predictions, for false and misleading statements about their use of AI; combined penalties of $400,000[8] | Claims about AI capabilities are held to the same standard as any other representation |
| June 2025 | Withdrawal of the 2023 proposal on conflicts of interest from predictive data analytics, along with other pending proposals[9] | No AI-specific rule is pending; existing obligations govern |
| FY2026 | Examination priorities list AI under emerging financial technology[7] | AI use is reviewed in exams now |
The SEC's Division of Examinations says it will review registrants' representations about their AI capabilities for accuracy, and assess whether firms "have implemented adequate policies and procedures to monitor and/or supervise their use of AI technologies."[7] The withdrawal of the 2023 proposal removed a possible AI-specific rule, but it didn't remove the scrutiny. The practical result is the same as in Canada: existing obligations, applied to new tools, tested in exams.
Read together, the guidance from both countries is consistent. For firms using AI to research and monitor products, it comes down to a short list of expectations.
| Theme | Canada | United States |
|---|---|---|
| Existing rules apply | Securities laws are technology-neutral (CSA 11-348) | FINRA rules are technology neutral and continue to apply (RN 24-09) |
| Governance | Governance and risk management "paramount"; document AI use (CSA, CIRO); model risk management for bank groups (OSFI E-23) | Formal review and approval of use cases; governance or model risk framework (FINRA) |
| Testing | Fit for purpose; tested before and after adoption (CSA) | Evaluate before deploying; robust testing (FINRA) |
| Human oversight | Human in the loop; output treated as an input, not a decision (CSA) | Human-in-the-loop review; caution on autonomous agents (FINRA) |
| Ongoing oversight | Evidence of ongoing oversight of algorithmic models (31-368) | Ongoing monitoring of outputs (FINRA); supervision of AI use reviewed in exams (SEC) |
| Explainability and records | Explainability sufficient to meet record keeping requirements (CSA) | Auditability a stated risk for AI agents (FINRA) |
| Third-party tools | Vendor models within scope of model risk management (OSFI E-23) | Rules apply to third-party and embedded tools (FINRA) |
| Accurate claims | AI washing may be a misrepresentation (CSA) | Enforcement for misleading AI claims; representations reviewed in exams (SEC) |
The one line no regulator in either country has crossed is accountability. None has suggested that using an AI tool changes who is responsible for understanding a product. The firm approves the product; the advisor understands what they recommend; the AI helps both do that faster.
| Use | What the AI Does | Main Risk | Control |
|---|---|---|---|
| Document extraction | Pulls fees, terms, parties and risk factors from prospectuses, offering memoranda and term sheets | Wrong or missing values presented confidently | Every extracted value linked to its source page; sample checked by a person |
| Summarization | Drafts product summaries and change notices | Plausible text that omits or misstates a risk | Human review and sign-off before issue; summary cites its sources |
| Change detection | Compares new filings and data against the product's file and flags differences | Missed changes; silent failures when data is unavailable | Tested against known changes; failures reported, not skipped |
| Scoring and screening | Ranks or scores products against criteria | Opaque results the firm can't explain | Documented methodology; outputs explainable to a reviewer and an examiner |
| Research assistance | Answers questions about a product for analysts or advisors | Unsupported or invented answers | Answers grounded in and linked to source documents; treated as an input only |
| Autonomous agents | Takes multi-step actions, such as updating records or statuses | Acts beyond scope; hard to audit | Narrow permissions; no status or approval changes without a person; full action log |
Sources are the control that runs through all of these. When every AI output can be traced to the document, page or data point it came from, a reviewer can check it, an advisor can rely on it, and an examiner can follow it. When it can't, the firm is left defending text it can't verify.