Client Consent for AI Meeting Recording in Canada: A Three-Layer Guide for Advisors

Do I need client consent to record a meeting with AI in Canada?

Recording a client meeting with an AI note-taker raises two Canadian consent obligations. Criminal Code section 184 makes the recording lawful when one party consents. PIPEDA requires the client's meaningful consent to the collection, use, and AI processing of the personal information in the recording. Both apply on every recorded meeting, and CIRO's recordkeeping rules apply to the resulting records.

Canadian advisors often hear their country described as a "one-party consent" jurisdiction, meaning one participant in a conversation can lawfully record it without telling the other. That framing is accurate for the Criminal Code. It covers one layer of the stack. The Criminal Code addresses whether the recording itself is lawful. PIPEDA governs what the advisor can then do with the personal information the recording contains. CIRO's books-and-records rules apply once that information becomes a firm record. Each layer has its own statute and its own duties. Consent to record, under the Criminal Code, covers only that layer.

The three-layer consent stack every Canadian advisor needs to know

Three separate layers of Canadian law apply to a recorded AI-assisted client meeting. I'm calling it the three-layer consent stack. This is a synthesis of the underlying law, rather than an official CIRO framework. Each layer has its own question, its own statute, and its own duty on the advisor.

1. Layer 1: The act of recording. The Criminal Code governs whether the recording itself is lawful. Section 184(1) prohibits the interception of private communication; section 184(2)(a) creates a one-party consent exception when a participant in the communication agrees.
2. Layer 2: The handling of personal information. PIPEDA governs what can be collected, used, disclosed, and processed, including by an AI tool. The client's consent must be meaningful under the OPC's 2018 Guidelines, which set out elements the advisor should name up front.
3. Layer 3: What the recording becomes once it enters the firm. CIRO's IDPC rules govern books-and-records, supervision, retention, and outsourcing obligations once the recording or the AI-generated summary becomes a firm record.

Criminal Code section 184 and one-party consent

One-party consent in Canada means that one participant in a private communication can lawfully record it even where the other party has not been told. Criminal Code section 184(2)(a) creates this exception to the general rule in section 184(1). A CIRO-registered advisor who takes part in a client meeting is a participant for this purpose.

The Supreme Court confirmed the one-party rule in Goldman v. The Queen in 1979. The majority held that consent by one party, whether originator or intended recipient, makes the recording lawful under what is now section 184(2)(a). The consent must exist at the moment of recording.

Turning on an AI note-taker on your own device during a client Zoom call satisfies section 184. You are a participant, and your own consent covers the interception. Every other consent question is still open. If your assistant starts the recorder before you join the call, the assistant is outside the call. With that in mind, the recording is an unlawful interception.

A note on Duarte

Advisors who have heard of Canadian recording case law have usually heard of R. v. Duarte (1990). The case is often paraphrased as the Supreme Court saying judicial authorization is needed to record a private conversation. Duarte involved an undercover police officer; the Supreme Court held that state-actor recording engages section 8 of the Charter and requires prior judicial authorization. A financial advisor recording a client meeting is a private party. The Charter analysis in Duarte applies only to state actors.

Why notes still work

In R. v. Fliss (2002), the Supreme Court excluded a covert police recording under section 24(2) of the Charter but held that the officer's independent oral testimony about the same conversation was admissible even without the tape. Memory sits outside the statutory definition of intercepted communication. An advisor's careful contemporaneous notes carry the same independent evidential weight.

PIPEDA meaningful consent for AI processing

PIPEDA requires meaningful consent before client personal information is collected, used, or disclosed, including when an AI tool processes it. The OPC's 2018 Meaningful Consent Guidelines name four key elements the advisor must cover up front: what information is collected, with whom it will be shared, the purposes, and the risks of harm.

The Criminal Code addresses whether the recording is lawful. PIPEDA governs what happens to the personal information inside the recording. These are separate consent questions, with separate standards.

The trigger stack

PIPEDA Principle 4.3.4 allows consent to be implied or express depending on the circumstances. The OPC's Meaningful Consent Guidelines name three triggers that generally require express consent: sensitive information, collection or use outside reasonable expectations, or meaningful residual risk of significant harm. All three fire for advisor AI meeting recording.

What clients must actually be told

Meaningful consent under Principle 4.3.2 requires the client to receive the information in a form they can reasonably understand. The Guidelines name four elements the advisor must cover up front:

• What is collected. Audio of the meeting, a transcription, and any data derived from them.
• With whom it will be shared. The vendor, any sub-processors, and the geographic location of processing.
• Purposes. Summarization, CRM record creation, compliance documentation, and whether data trains the vendor's models.
• Risks. Inaccurate summaries in regulatory documentation, vendor retention, cross-border transmission, and residual harm from vendor breach.

Withdrawal, and the honest tension with CIRO retention

Principle 4.3.8 gives clients the right to withdraw consent at any time, subject to legal or contractual restrictions. PIPEDA Case Summary #2003-249 held that informing a customer of the right is insufficient when the operational mechanism does not work. For AI recording, the operational obligation requires the advisor to stop mid-meeting on request, delete the audio, transcript, embeddings, and vendor-side derived data, and confirm deletion back to the client.

A genuine tension sits between Principle 4.3.8 and CIRO recordkeeping. A client may withdraw consent after the meeting summary has become firm books-and-records material. The resolution is asymmetric: the regulated summary stays; the supporting AI artifacts come out. Firm counsel is the right place to confirm which records belong in which bucket.

CIRO supervision once the recording becomes a record

CIRO has not published a rule that prescribes a particular form of client consent for meeting recordings. Once the recording or an AI-generated summary enters the firm's books and records, CIRO recordkeeping, supervision, and retention rules apply. The consent itself sits under PIPEDA and, where applicable, Quebec Law 25, BC PIPA, or Alberta PIPA.

The four buckets

Records. An audio recording, a transcript, or an AI-generated summary becomes a firm record when its content relates to the firm's business or a client transaction. IDPC Rule 3800 applies medium-neutrally; MFD Rule 5 applies in parallel for mutual fund dealers. Seven-year retention runs from creation.

Supervision. Where a recording or summary informs a suitability decision, KYC update, or trade instruction, it joins the supervisory file under IDPC Rule 3900.

Outsourcing. The AI vendor is a third-party service provider. GN-2300-21-003 requires written contract, documented due diligence, regulator access rights, and ongoing monitoring. Vendor failure is firm failure.

Material change. Deploying a new AI tool may itself be a material change to the firm's business that requires notification to CIRO. The 2026 Annual Compliance Report explicitly directs dealers to make this assessment.

What credible documentation looks like

CIRO's Re White (2024 CIRO 67) considered handwritten contemporaneous notes. The panel found the notes "tip the balance strongly in favour of Mr. White's narrative" (para. 125). What transfers to an AI workflow is the credibility architecture the panel applied, by analogy rather than by holding. The architecture has three parts: contemporaneity, custody, and corroboration.

Where recording consent ends and processing consent begins

The most common consent script in advisor practice is a single line at the start of the meeting: "I'm recording this, is that OK with you?" If the client says yes, the advisor moves on. That single yes handles one Canadian obligation. It does not handle the other.

What changes when you cross the provincial line

The three-layer stack applies everywhere in Canada. What changes by province is whether additional obligations sit on top.

When the client says yes to the meeting but no to the AI

Client responses cluster into three patterns: a clean no, a mid-meeting change of mind, and a yes-with-limits. Each has a different next move for the advisor.

If the client says no

If the client says no to the AI tool, the meeting proceeds without it. The client's refusal has no regulatory cost to them and no documentation gap for the advisor, provided non-AI notes meet the contemporaneity, custody, and corroboration standard.

If the client changes their mind

If the client asks partway through to stop recording, the advisor stops. PIPEDA Principle 4.3.8 treats withdrawal as operational: the recording ends, and anything captured from that point forward is not retained.

If the client says yes to some but not all

If the client is comfortable with the advisor recording for note-taking purposes but not with the transcript going to a third-party AI vendor, the request is coherent under the two-consent structure. What the advisor cannot do is assume the first yes covers the second question.

What to say: practical consent scripts

The three-layer stack collapses in practice into a consent conversation that takes under a minute. The three scripts below handle new clients, existing clients starting to use an AI tool mid-relationship, and meetings where the question comes up after the conversation is already underway.

Script A: With a new client

Script B: With an existing client

Script C: Mid-meeting

What to write down: the CRM audit trail

The record is what an examiner reconstructs. For AI meeting recording, ten categories of CRM fields together build the audit trail that shows consent was obtained, documented before processing began, and honoured through withdrawal and retention.

1. Identity and versionConsent form version, Quebec client flag, AI tool version.
2. Consent eventSeparate decisions for recording, AI processing, and cross-border disclosure.
3. Meeting and processing timestampsIncluding Consent_Confirmed_Before_Processing.
4. Data location and routingPrimary storage region, inference routing jurisdictions.
5. Withdrawal and partial consentScope of consent, scope of refusal, mid-meeting changes.
6. Evidence artefactsSigned PDF, consent form template version, privacy policy version.
7. Retention specificationSeven-year CIRO minimum, five-year Quebec incident register.
8. Workflow integrationWhen each field is populated across the meeting lifecycle.
9. Reconstruction bundleA nine-item evidentiary bundle for regulator review.
10. Refusal and withdrawalWhat to document when a client declines or withdraws.

What US vendors tell you, and what they can't

US AI vendors market to a landscape where wiretap law varies by state, privacy statutes are fragmented, and "no recording" can be a marketing advantage. None of those frames translate cleanly to Canada.

"No recording, no consent question"

Some AI vendors market products that process audio in real time without persisting a recording file. The framing implies that if no recording exists, no consent question arises. That framing does not hold in Canada. Criminal Code section 184 prohibits the interception of a private communication, and "intercept" is defined in section 183 to include listening to a communication, not only recording it.

Brewer v. Otter.ai and US litigation

Brewer v. Otter.ai (US, Northern District of California, filed August 2025) is a class action alleging Otter recorded meetings without all parties' consent. US wiretap statutes do not bind Canadian practice. What the filing signals is a shift in how regulators and courts view AI meeting tools.

What to do with this

Three-layer consent looks heavier on paper than it has to feel in practice. A one-minute conversation at the start of the meeting, a written acknowledgement on file, and a CRM record that shows consent preceded processing are the working parts. Each layer of the stack carries its own regulator, but the advisor's daily handling is one conversation, one document, and one field in the CRM.

Canadian regulators are moving toward a position where AI-assisted documentation is expected to meet a higher bar than traditional notes, not a lower one. The advisor who builds the three-layer discipline now is positioning for the exam before the rule arrives.