Anonymised, illustrative composite. A client’s divorce-driven sale details ended up in a casual conversation with someone who had no business knowing them — and the trail led back to an AI note-taking tool nobody had actually vetted.
At a glance
A seller in the middle of a divorce engaged an agent to sell quickly and quietly — explicitly asking that details about the sale, and the reasons for it, stay between the two of them. On a planning call, the seller disclosed a firm floor price roughly 8% below list and their reasons for wanting privacy from the soon-to-be-former spouse’s wider circle.
To save time, the agent ran the call through a free, personal AI transcription and summarization tool bolted onto their phone app — not the brokerage’s vetted CRM, just a convenient add-on. The tool’s auto-generated summary synced, by default, into a shared team workspace two licensed assistants and a marketing coordinator all had standing access to, none of whom had any reason to see this particular file. About three weeks later, the client heard the floor price mentioned, almost in passing, by an acquaintance with no connection to the transaction — a detail that had existed nowhere except that private call and the AI-generated note built from it.
One 42-minute call, one AI-generated summary, shared by default with three people who did not need it, and a sensitive detail — a discounted floor price the client had asked to keep confidential — surfacing outside the transaction within roughly three weeks of the call. Nothing about the leak required malice; a default sharing setting did all the work on its own.
PIPEDA’s test for meaningful consent requires an organization to make clear what information is collected, with whom it is shared, for what purpose, and the risk of harm involved — and requires express, not merely implied, consent wherever the information is sensitive or the handling falls outside what the person would reasonably expect. Financial position and a client’s reasons for selling during a divorce are squarely sensitive; routing that conversation through an unvetted third-party AI tool, and sharing the resulting note with staff who had no role in the file, is exactly the kind of disclosure the reasonable-expectations test exists to catch. Nobody asked the client’s permission for either step.
CREA’s own guidance draws the same line without softening it: “the use of technology, such as AI, does not alter a REALTOR®’s obligations” — a member remains fully responsible for information handled on a client’s behalf whether or not an AI tool touched it. RECO’s own confidentiality rule sits on top of that: a brokerage “must ensure that administrative or support staff… who might have access to confidential client information… do not share the client’s information,” and any consent to share confidential information must specify what, with whom, for what purpose, and to whose benefit — none of which existed here.
The managing broker required the agent to stop using the personal AI tool for any client call, move exclusively to the brokerage’s vetted CRM feature (contractually confirmed not to train on client data or leave the brokerage’s systems), obtain express written consent before any AI-assisted note-taking on a call touching sensitive matters, and restrict internal note access to staff with an actual reason to see the file. The client accepted the corrective steps once they were explained and did not escalate further.
PIPEDA defines significant harm broadly enough to reach this fact pattern directly — humiliation and damage to relationships are named examples, and a divorce-adjacent financial disclosure sits well inside that category. Had the affected client pursued it as a breach, the real-risk-of-significant-harm test could plausibly have been met, triggering mandatory notification to the Privacy Commissioner and to the client “as soon as feasible,” plus a mandatory 24-month record of the incident regardless of the outcome. Separately, knowingly failing to meet either of those duties carries its own penalty under the Act: up to $10,000 on summary conviction, up to $100,000 on indictment — the cost of mishandling the response, layered on top of whatever the underlying disclosure itself already cost the client’s trust.
Any workflow where a client call about a sensitive personal matter — a divorce, a job loss, a financial hardship — runs through a tool nobody has separately reviewed, landing by default in a shared space, is a leak waiting to surface. The giveaway is usually that the sharing was never a deliberate choice by anyone; it was simply the tool’s default behaviour, and nobody checked what that default actually was before the first sensitive call went through it.
Related reading: meaningful consent, defined express consent, defined personal information, defined
A 30-minute call is enough to tell you whether AI pays for itself here.
