Territory 5 · AI governance
Can Canadian Financial Advisors Use Claude? A Compliance Analysis
Claude reaches an advisor by three kinds of route, and the terms differ on each. Here is what each means under PIPEDA, CIRO, and Quebec Law 25.

If you have searched for guidance on using AI as a Canadian financial advisor, much of what you found is likely to be about ChatGPT. Claude, built by Anthropic, comes up less often. And yet British Columbia Investment Management Corporation has presented its use of Claude for investment research and financial modelling in an Anthropic webinar (external site, opens in a new tab). An article on RBC Capital Markets' site, published 14 May 2026 (external site, opens in a new tab), describes Aiden, RBC's enterprise AI platform, as "built in partnership with Nvidia and Anthropic". Sun Life runs an internal chatbot on Claude that handles over 10,000 employee queries a week, according to an AWS case study (external site, opens in a new tab).
The institutional interest is real. Adoption by individual advisors is much less visible. When this page was first published, on 15 April 2026, Northern Catalyst had found no documented case of a CIRO-registered advisor using Claude directly for client work.
That contrast is the reason for this page. Claude's terms differ from ChatGPT's in ways that matter for Canadian advisors. If you have already read the companion ChatGPT compliance analysis, the regulatory framework (PIPEDA, CIRO, Quebec Law 25) carries over. This page focuses on what is different when the tool is Claude.
The short answer: Canadian advisors can use Claude for general research, drafting, and brainstorming, just as they can with ChatGPT. But Claude reaches an advisor by three kinds of route, and the terms differ on each. The terms for the individual plans also carry a restriction that names securities. Understanding which route you are on, and what that means under Canadian law, is the work this page does.
Three kinds of route, three sets of terms
Which terms apply depends on the route. As at 13 November 2026, Anthropic draws a line between use by individuals and commercial use. On one side of that line, its Consumer Terms of Service (external site, opens in a new tab) apply and conversations may be used to train Anthropic's models unless the user opts out. On the other side, its Commercial Terms of Service (external site, opens in a new tab) apply: Anthropic is the customer's data processor and may not train models on customer content unless the customer opts in to a program that allows it. On the third kind of route, a cloud platform such as Amazon Bedrock, Anthropic's documentation (external site, opens in a new tab) says the cloud provider is the data processor.
Claude's plans for individuals (Free, Pro, Max, and Ultra)
Anthropic's Consumer Terms of Service (external site, opens in a new tab) (effective 8 October 2025, and still the version in force as at 13 November 2026) allow Anthropic to use conversations to train its models "unless you opt out of training through your account settings". Anthropic's announcement of 28 August 2025 (external site, opens in a new tab) states that when a user turns the setting off, their data "will still be included in model training that has already started and in models that have already been trained".
Paying for a plan leaves all of this as it is. The Consumer Terms cover Anthropic's services for individuals, which as at 13 November 2026 are the Free, Pro, Max, and Ultra plans (external site, opens in a new tab).
There is a further detail in Claude's terms. Section 3 of the Consumer Terms lists, among uses that are not permitted, relying on the services "to buy or sell securities or to provide or receive advice about securities, commodities, derivatives, or other financial products or services", and gives as its reason that Anthropic is not a broker-dealer or a registered investment adviser. Northern Catalyst reads this as a disclaimer more than a practical barrier to meeting documentation: an advisor summarizing a conversation is using Claude for writing, not for investment guidance. That is one reading of broadly worded terms, and whether a given use falls within the clause is for the firm's compliance officer to decide.
As at 13 November 2026, Anthropic's Privacy Policy (external site, opens in a new tab) (effective 10 September 2026) states that personal data "is transferred to our servers in the US, or to other countries", and its supplemental disclosures for residents of Canada state that personal data may be processed and stored "outside of Canada". No Canadian data residency option is offered on the individual plans. Anthropic's Data Processing Addendum (external site, opens in a new tab) belongs to its Commercial Terms, which do not cover those plans.
A note on Claude's desktop app and Cowork. A desktop app can give a false sense of locality. As at 13 November 2026, Anthropic's help centre (external site, opens in a new tab) states that Cowork runs its tasks in the cloud: "Claude's work runs on Anthropic's servers, in an isolated environment, and your sessions and files are saved to your Claude account." An internet connection is required throughout. A meeting transcript opened in Cowork leaves the device, whether the app is on a desktop, in a browser, or on a phone. What governs the data is the plan's terms (individual, Team, or Enterprise), not the form of the app.
A note on Claude within Microsoft 365. Microsoft announced Copilot Cowork (external site, opens in a new tab) on 9 March 2026 and made it generally available (external site, opens in a new tab) on 16 June 2026, stating that at general availability it "runs on Anthropic models". This can create a different false assumption: that "it's in my Microsoft tenant, so it's covered by my Canadian data residency." For a tenant whose sign-up country is Canada, Microsoft commits to store (external site, opens in a new tab) the content of Microsoft 365 Copilot interactions at rest in Canada. Claude models are treated differently. As at 13 November 2026, Microsoft states (external site, opens in a new tab) that Claude models in its Copilot offerings "are currently excluded from the EU Data Boundary, and when applicable, in-country processing commitments", and that it "enables Claude models on by default for most customers in commercial cloud (excluding EU/EFTA and UK)". The setting sits in the Microsoft 365 admin center under Copilot, Settings, "AI providers operating as Microsoft subprocessors". A firm that has data residency expectations will want its administrator to have looked at it.
Under PIPEDA, the reasoning in the companion ChatGPT compliance analysis applies: typing client information into a consumer AI tool likely constitutes a "disclosure" to a third party for its own purposes, which is the kind of disclosure for which the Office of the Privacy Commissioner of Canada (OPC) said express opt-in consent should have been obtained in its Home Depot findings (external site, opens in a new tab) (PIPEDA Findings #2023-001, 26 January 2023).
Claude API and commercial plans (Team and Enterprise)
The terms are different here. As at 13 November 2026, under the Commercial Terms of Service (external site, opens in a new tab), Anthropic is the customer's data processor and "may not train models on Customer Content" unless the customer opts in to a program that allows it. A Data Processing Addendum (external site, opens in a new tab) (DPA) that includes the EU Standard Contractual Clauses is incorporated into those terms. The securities restriction in Section 3 of the Consumer Terms does not appear in the Commercial Terms.
As at 13 November 2026, Anthropic's privacy centre (external site, opens in a new tab) states that for API users it deletes inputs and outputs on its backend "within 30 days of receipt or generation", with listed exceptions. OpenAI's API documentation (external site, opens in a new tab) states that abuse monitoring logs are "retained for up to 30 days" by default. Both offer zero data retention arrangements to customers they approve.
As at 13 November 2026, Anthropic's API stores data at rest (external site, opens in a new tab) in the United States, the only storage location it offers. By default the model's computation "may run in any available geography"; a customer can restrict it to the United States. No Canadian option exists for storage or for processing. What this route adds over the individual plans is contractual: the processor relationship, the Data Processing Addendum, and the default bar on training.
Claude through Amazon Bedrock (data at rest in Canada)
Amazon Bedrock runs Claude models inside AWS. As at 13 November 2026, Anthropic's Service Specific Terms (external site, opens in a new tab) state that, subject to a carve-out for the models Anthropic designates as Covered Models, "the technology provided by Anthropic to enable access to the Hosted Services does not give Anthropic access to any Customer instance within the applicable Cloud Platforms, including the Inputs or Outputs they contain." Anthropic's documentation (external site, opens in a new tab) lists the Covered Models on that date as Claude Fable 5, 5.1, and 5.2 and Claude Mythos 5, 5.1, and 5.2, and for those models the terms say Anthropic "may retain and perform safety reviews on Inputs, Outputs, and other data". AWS's documentation (external site, opens in a new tab) states that model providers "don't have access to Amazon Bedrock logs or to customer prompts and completions."
As at 13 November 2026, Amazon Bedrock is available in two Canadian Regions, ca-central-1 (Canada Central) and ca-west-1 (Canada West, Calgary). AWS states (external site, opens in a new tab) that data at rest, "including logs, knowledge bases, and any stored configurations", stays in the Canadian Region the customer uses.
There is an important qualification: where a request is processed depends on the model and the inference profile. On 27 October 2026 AWS announced (external site, opens in a new tab) Canada cross-Region inference for Claude Haiku 5.5, which "routes requests only between ca-central-1 (Canada Central) and ca-west-1 (Canada West, Calgary)", so that customers "can now keep inference processing entirely within Canada." AWS describes it as "an inference profile with limited model selection". As at 13 November 2026 that announcement names no other Claude model. On the same date AWS's Region table (external site, opens in a new tab) shows Claude Fable 5.2 and Claude Mythos 5.2 with no in-Region support in either Canadian Region; from Canada they are reached through cross-Region inference, under which AWS can route a request to a Region outside Canada. AWS's advice (external site, opens in a new tab) is: "If you have strict data residency or compliance requirements, you should carefully evaluate whether cross-Region inference aligns with your policies and regulations".
What this means in practice: under United States or global cross-Region inference, the text of a client meeting summary is processed outside Canada and returned over AWS's own encrypted network. For most models AWS states (external site, opens in a new tab) that no customer data is stored in the Region that does the processing. For the Claude Fable 5 models, AWS states (external site, opens in a new tab) that retained prompts and completions "are stored in destination regions", that is, where the request was processed (as at 13 November 2026).
As at 13 November 2026, what Amazon Bedrock keeps depends on the model. AWS states (external site, opens in a new tab) that for Claude Fable 5, Claude Fable 5.1, and Claude Fable 5.2, "user prompts and completions are retained within the AWS boundary for up to 30 days and may be reviewed by AWS", and that "Your content is not shared with the model provider." For earlier Claude models AWS states: "There is no data retention change to Claude models released before Claude Fable 5." Anthropic's terms and AWS's page describe that review in different words, so a firm weighing this route will want to read both.
What Canadian regulators need you to know
The full Canadian regulatory framework for AI tools (PIPEDA accountability, CIRO supervision, Quebec Law 25) is covered in depth in the companion ChatGPT compliance analysis. The same framework applies to Claude. Three points bear on Claude in particular.
The OPC's findings of 6 May 2026 are about OpenAI. The joint federal and provincial investigation of OpenAI, opened in 2023, ended with findings published on 6 May 2026 (external site, opens in a new tab): the complaint was well founded and, under PIPEDA, conditionally resolved. That investigation focused on ChatGPT and the models behind it at the time, and did not extend to how OpenAI's clients, such as API customers, use the tool. They are findings about OpenAI and its models, not about Anthropic or Claude.
What each provider publishes on Canadian privacy law. As at 13 November 2026, Anthropic's Privacy Policy (external site, opens in a new tab) (effective 10 September 2026) carries supplemental disclosures for residents of Canada and does not name PIPEDA. Its Data Processing Addendum (external site, opens in a new tab) refers to "Applicable Data Protection Laws" in general terms and includes the EU Standard Contractual Clauses, a UK Addendum, and a Swiss Addendum, with no Canada-specific addendum. OpenAI's Data Processing Addendum (external site, opens in a new tab) (effective 1 January 2026) likewise has no Canada-specific terms. Google Cloud publishes a PIPEDA whitepaper (external site, opens in a new tab), and Microsoft publishes a Canada privacy laws page (external site, opens in a new tab) for Azure that points to a privacy impact assessment covering PIPEDA. A firm evaluating Claude or ChatGPT does its own PIPEDA analysis on top of these general contractual terms.
CIRO's examination approach applies whichever tool is used. CIRO's Compliance Report for 2026 (external site, opens in a new tab) (Administrative Bulletin 26-0034, released 17 February 2026) says: "As part of our FinOps examination approach, we will be enquiring about the use of AI in dealers' operations. To the extent dealers are using AI, we will be reviewing the operational controls they implemented to ensure AI is working as designed." That section of the Report names no AI provider or product, so it applies to Claude as it does to any other tool. What it means for meeting notes is covered in Does CIRO Allow AI Meeting Notes?.
For advisors in Quebec, Law 25 applies as well. Section 17 (external site, opens in a new tab) requires a privacy impact assessment before personal information is communicated outside Quebec. The Act draws the line at Quebec's border, not Canada's, so it reads as covering a transfer to another province as well as to another country. That holds whichever AI provider receives the information. Law 25 may also apply to advisors elsewhere who serve clients resident in Quebec. The companion ChatGPT compliance analysis covers Law 25 in more detail.
How the terms for Claude and ChatGPT compare
Northern Catalyst built its own documentation tooling on Claude through Amazon Bedrock. Given that, this comparison keeps to what each vendor's own pages say.
On the business routes (API, Enterprise, or cloud-hosted), Anthropic and OpenAI publish similar commitments as at 13 November 2026. Both state that they do not train on business customers' data by default (Anthropic (external site, opens in a new tab), OpenAI (external site, opens in a new tab)). Both offer Data Processing Addenda. Both list SOC 2 Type II, ISO/IEC 27001, and ISO/IEC 42001:2023 (Anthropic (external site, opens in a new tab), OpenAI (external site, opens in a new tab)); OpenAI also lists ISO/IEC 27701.
The cloud-hosted comparison moves quickly. As at 13 November 2026: on Amazon Bedrock, AWS states that content is not shared with the model provider, offers a Canada-only inference profile for Claude Haiku 5.5, and keeps prompts and completions for up to 30 days for the Claude Fable 5 models. On Azure, Microsoft states (external site, opens in a new tab) that under a Standard (regional) deployment "Prompts and responses are processed within the customer-specified Azure geography"; its availability table (external site, opens in a new tab) lists regional Standard deployments in Canada East, with gpt-4o and gpt-4.1-mini, and none in Canada Central. Both vendors revise these tables often, so the table itself is what to read on the day a decision is made.
Both providers are companies subject to United States law, including the CLOUD Act provision at 18 U.S.C. s. 2713 (external site, opens in a new tab), which reaches data in a provider's "possession, custody, or control" wherever it is stored.
In summary: neither platform is ahead on every count. Which properties matter depends on the firm's priority: where the model's computation happens, how long content is kept, and who can see it. Each vendor's current table answers those questions, and the answers have moved more than once in 2026.
One design approach: remove personal information first
Some tools address the cross-border question through design: personal information is detected and replaced with placeholders on the advisor's own device before a transcript is sent for processing, and restored locally when the summary returns.
Where the model's processing happens outside Canada, as described above, the text still crosses the border. What crosses it has the detected names, account numbers, and addresses replaced, so the question for PIPEDA purposes moves from "did data leave Canada?" to "what kind of data left Canada?" The design has a limit: it replaces what it detects, and a detailed transcript with the names removed can still point to a person, as the page on meeting transcription sets out.
Northern Catalyst used this design in documentation tooling it built on Claude through Amazon Bedrock. The principle holds regardless of which AI model sits behind it.
A second approach keeps the work inside systems the firm already governs. Northern Catalyst's Advisor Copilot Toolkit takes that route: every agent is built inside the firm's own Microsoft 365 tenant and runs there. Those agents do not remove client personal information, and processing, permissions, policy, and data-location terms are set by Microsoft and the firm's administrator. The note on Claude within Microsoft 365 above applies to that route.
What to do now
The steps are the same ones outlined in the companion ChatGPT compliance analysis, adapted for Claude.
First, check which route you are on. If you use claude.ai with a personal account (Free, Pro, Max, or Ultra, as at 13 November 2026), the Consumer Terms apply. Your conversations may be used for training unless you opt out. No data processing agreement stands between you and Anthropic. For general research and drafting that involves no client information, this is fine. For anything involving client data, it is not.
Second, talk to your compliance officer. The 2026 Compliance Report says CIRO's examiners will be asking about the use of AI in dealers' operations. If your firm has no published AI use policy, that conversation is worth having whichever tool you prefer.
Third, for client-specific work, use a tool your firm has approved, under an agreement that covers where client information is processed, how long it is kept, and who can see it. The four questions in Getting Started with AI as a Canadian Financial Advisor apply to any tool.
At the end of the path
Free. The material for the conversation with the compliance officer.
This is educational content based on published regulatory sources as at the page's last-updated date. It is not legal, compliance, or financial advice. Review specific compliance decisions with your compliance officer and qualified legal counsel before implementation.