The CEO described advisers enabling separate AI notetakers during a recurring internal call, then allowing the services to distribute summaries after the meeting. The immediate result was an inbox full of redundant, slightly different accounts of the same discussion. The proposed fix—designate one or two official notetakers and tell everyone else to switch theirs off—is sensible, but it understates the operational failure that allowed the tools into the meeting in the first place.
Advisor.ca did not name the wealth manager, the meeting platform, the AI products involved, or whether the tools were firm-approved. It also did not say whether those summaries were retained in corporate Microsoft 365 tenants, transmitted by third-party vendors, or generated from recordings held outside the firm’s managed environment. Those omissions matter. A harmless internal nuisance and an uncontrolled disclosure of regulated business communications can look identical at the inbox level.
Twenty summaries mean twenty records to control
The most important fact in the account is not the number of email notifications. It is that every AI assistant may have made its own recording or captured its own audio stream, created a transcript, extracted action items and sent a derivative summary to one or more people. Even if all 20 summaries were broadly accurate, they are not necessarily interchangeable records.
A traditional meeting-minutes process has an owner. Someone takes notes, an accountable person approves them, the final version is stored in a known location, and the draft material is handled under a defined retention rule. AI notetakers reverse that discipline when staff activate them individually. The organization may end up with several conflicting descriptions of a discussion, no recognized authoritative version and no straightforward way to determine who received what.
That is particularly awkward for wealth-management firms. Communications with clients and records relating to firm business can carry regulatory retention and supervision duties. CIRO has long treated client business communications as firm records that must be maintained. If an adviser uses a personal transcription service, then sends its output into a personal inbox or an unapproved cloud workspace, the firm has a discoverability and supervision problem before anyone even assesses whether the summary was correct.
The failure is easy to miss because the AI tool presents itself as a productivity feature. A bot joining a Teams, Zoom or Google Meet call looks like an attendee. A post-meeting recap looks like a convenience. But the service is also creating a new information system around a meeting: it has identities, permissions, data storage, deletion controls, exporting features and its own security terms. Treating that as mere etiquette is how unmanaged shadow IT becomes routine.
For Windows administrators, the practical lesson is straightforward: do not make “please turn off your bot” the primary control. Use platform settings and conditional access rules where available, limit who can install or connect third-party meeting applications, and decide which transcription service—if any—is approved for each meeting category. A policy that rests on every participant remembering to disable their personal assistant will work until the first calendar integration automatically joins one anyway.
One designated transcript needs a designated owner
The better policy is not a blanket ban on transcription. An accurate transcript and concise summary can improve recall, accessibility and accountability, especially during internal operations meetings. The issue is that the firm must determine the tool, the purpose, the people entitled to access the output and the system of record before the meeting begins.
A controlled approach separates meeting types. Routine internal training may be suitable for an approved Teams transcription workflow. A call involving client financial information, prospective transactions, account instructions, compensation or legal advice deserves a higher bar. In some cases, an approved human note-taker and a carefully reviewed note entered into the CRM may be safer than a full audio recording and machine-generated transcript.
The official meeting invite should make the rule visible rather than burying it in a general AI policy. It should identify whether recording or transcription will occur, which service is doing it, who will receive the notes, how long the source recording and transcript will be retained, and whom to contact if an attendee objects. Meeting organizers need authority to remove unexpected bot attendees and to pause discussion until the participant list is understood.
There is a significant difference between a Teams-generated transcript governed by a firm’s Microsoft 365 configuration and a standalone assistant connected through an employee’s personal account. IT teams should verify that distinction instead of assuming an app branded as an “AI meeting assistant” operates inside the organization’s tenant or follows its Microsoft Purview retention and eDiscovery policies.
The Advisor.ca account also exposes another weakness of AI summaries: duplication increases divergence. A summary is an interpretation, not a neutral recording. Different systems may identify different action items, assign comments to the wrong speaker, omit qualifiers or phrase uncertainty as a decision. When a CEO receives 20 summaries from the same call, the noise is irritating. When the subject is client suitability, a trade instruction or a compliance escalation, it creates a more serious question: which version, if any, was reviewed and adopted by the firm?
The answer should be explicit. The owner of the official transcript should review it, correct material errors, label it as the approved record where appropriate and store it in the prescribed location. Everything else should be prohibited or deleted under a documented process. “AI-generated” is not a useful excuse for an inaccurate record after it has been circulated internally or acted upon.
Smart glasses raise the consent bar
The opinion also points to Meta Ray-Ban smart glasses as a possible way for advisers to record offsite client meetings. This is where the etiquette argument becomes inseparable from privacy practice. A client who can see a laptop screen saying “Recording” has at least some chance to understand what is happening. A client speaking to an adviser wearing ordinary-looking glasses may not know that audio, video or both could be captured.
Canada’s Office of the Privacy Commissioner says organizations recording customer calls should tell customers that recording is taking place, state the purpose and obtain meaningful consent. It also says organizations must limit retention and protect the information. The same principles make a useful baseline for in-person AI-enabled recording, even where the technology and the provincial legal context differ.
Consent cannot mean a vague warning that “AI may be used.” A client needs the operational facts: whether audio or video is being captured; whether a transcript and summary will be created; whether the data goes to a third-party processor; who can access it; whether it could be used to train or improve a service; how long it will be kept; and what alternative exists if the client declines. In a financial-advice relationship, a meaningful alternative should be as simple as taking conventional notes or using an approved audio-free workflow.
The report is right that the adviser should switch to regular glasses if the client is uncomfortable. The more important control is to decide in advance whether wearable recording is allowed at all. A firm should not leave that judgment to individual advisers making a real-time call in a café, a client’s home or a conference room.
There is also a recordkeeping consequence that the etiquette framing leaves largely unstated. If a wearable device produces a transcript that captures client instructions or material financial information, that output may become part of the firm’s business record. If the device is not approved, the firm could face the worst combination: a sensitive record exists, but it may sit in an unmanaged consumer account with uncertain retention and audit controls.
Profanity filters are a policy issue, not a safety feature
One of the lighter anecdotes in Advisor.ca involves a fintech executive who swore after tripping over a dog while dictating to an AI assistant, only for the assistant to respond that the language was inappropriate. The story is amusing, but it identifies a real governance problem: AI tools have different content-handling behavior, and employees often discover those differences only after deploying them in business workflows.
The opinion says Microsoft Copilot masks profanity with asterisks while services such as Otter transcribe speech verbatim. Microsoft does document harmful-content protections for Microsoft 365 Copilot, while Otter describes its service as speech-to-text transcription and warns users to review transcripts for accuracy. But the product-specific comparison should not be treated as a universal compliance rule. Behavior can differ by product, tenant policy, language, feature, version and whether the user is dictating, transcribing a recording or asking a generative assistant to summarize it.
For administrators, the question is not whether a tool permits profanity. It is whether the firm knows what the approved tool changes, omits, stores and sends. Asterisks can conceal a quoted client statement, while verbatim transcription can preserve language that an adviser did not intend to distribute. Either outcome may be inappropriate in a formal record depending on context.
Firms should test approved tools with realistic scenarios before rolling them out: multiple speakers, names and account terminology, interruptions, accented speech, sensitive instructions, profanity, off-topic conversation and requests to amend or delete a transcript. A short technical validation is more useful than a generic statement that the vendor has “responsible AI” safeguards.
The recurring lesson from the wealth-management CEO’s inbox is that AI etiquette needs enforceable defaults. Staff should not have to infer whether a meeting bot is acceptable, whether a client can be recorded through smart glasses, or whether a transcript is an official record. Those decisions belong in meeting settings, approved-app controls, data-retention configurations and short procedures employees can actually follow.
The next monthly call should produce one approved record—or none at all. It should not produce 20 competing accounts of the conversation and force the CEO to discover the firm’s AI policy through a cascade of email notifications.
References
- Primary source: advisor.ca
Published: August 7, 2026 at 6:23 PM UTC
Loading…
www.advisor.ca - Related coverage: learn.microsoft.com
Manage harmful content protection settings for Microsoft 365 Copilot Chat | Microsoft Learn
Learn how to set up a policy that enables users to disable harmful content protection in Microsoft 365 Copilot Chat as appropriate.learn.microsoft.com - Related coverage: support.microsoft.com
Transparency Note for Microsoft Copilot | Microsoft Support
Transparency Note for Microsoft Copilotsupport.microsoft.com