SHADOW AI DISCOVERY
Shadow AI Assessment for Canadian Firms
A short, evidence-based review that shows where AI use is visible in the evidence your environment already records. We tie activity to work identities, devices or paid subscriptions where the evidence supports it, and we call out what the telemetry cannot prove. We work your Microsoft 365 tenant, your network traffic records, your managed browsers and your expense claims, then hand you findings your privacy lead can act on.
What the review covers
Four surfaces. We choose the order at scoping based on effort, signal quality and what your environment already records. Each answers a different question, and none of them is complete on its own.
| Surface | What we establish |
|---|---|
| Microsoft 365 and Entra ID (work sign-ins and app permissions) | Which work identities have authenticated to AI services, and which OAuth app permissions (apps your users have authorised to reach work data) and enterprise apps carry standing access |
| Network and DNS egress | Which devices talk to AI endpoints from the corporate network, including personal accounts your tenant cannot see |
| Browser extensions and profiles | Which AI assistants are installed in the browser estate, and which of them can read page contents |
| Expense claims and cards | Which AI vendors appear in expense claims or card records. A recurring charge is a useful lead for sustained use; we confirm the account owner and business dependency before treating it as one |
Google Workspace tenants follow the same logic through the admin console security reports and third-party app access controls. If you want the reasoning behind the four surfaces before you commit to anything, the companion guide on shadow AI in Canadian SMBs sets out the method in full.
What you get
A findings document written for a privacy lead rather than an engineer. It names the evidence found, the surface it appeared on, the identity, device or subscription attribution available, and what the tenant configuration permits. If client or personal information may be involved, it records what is confirmed and what still needs interviews, vendor records or other evidence. Where client or personal information is implicated, it sets out the facts a breach assessment needs, without making the notification call for you.
Alongside it, a prioritised control list: tenant-level application consent, an approved-tool position, Conditional Access sign-in rules, and the data classes worth labelling first. Fusion Computing configures and verifies those controls where you want them implemented, as part of the managed IT work we already do.
How the engagement runs
We start with a scoping call to establish what your tenant can actually report on, which varies more than most firms expect. Read access to the relevant admin surfaces follows, and that access is time-bound, least-privilege and recorded, then the review itself, then a working session on the findings with whoever owns privacy in your firm.
The work is CISSP-led. We do not quote implementation before the findings exist, because the control list depends on what turns up.
Who this is for
Canadian firms between roughly 10 and 150 seats where confidentiality is a professional duty rather than a preference. That includes law firms, accounting practices, clinics, brokerages and wealth managers. It also fits owner-led businesses where staff adopted AI before the business made a decision.
It suits a firm that suspects the answer and wants it established properly. If the scoping call shows there is no credible shadow-AI question or useful evidence to review, we will tell you and stop there. The scoping call does not certify that no AI use exists.
What we need from your team
Read access to the Microsoft 365 or Google Workspace admin surfaces, a contact in finance who can search expense lines, and someone who can answer questions about which tools people were told they could use. Nothing needs to be installed on endpoints for the core review.
What it costs
Scope decides it. The drivers are seat count, the admin surfaces available, whether network traffic logging exists, and whether you want controls implemented or only documented. A single-tenant firm with existing logging is a materially smaller job than a two-tenant firm with none.
We price the review after the scoping call, once we know what your environment can report. You will have the number before any work starts.
How this fits with our other AI work
This review answers what is already happening. Our AI readiness assessment asks whether the firm is ready to adopt AI deliberately. It is the better starting point when current AI use is already understood and the goal is planned adoption. The two run in different directions, and discovery usually comes first, because readiness planning over an unknown estate is guesswork.
If what you need is the document rather than the discovery, our guide on what belongs in an AI acceptable use policy covers that, and the ready-to-fill LSO and CPA templates cover the regulated professions. Sector-specific duties are set out in our guides for law firms and accounting firms.
Frequently asked questions about shadow AI assessments
How long does a shadow AI assessment take?
Scope decides it. The admin-surface work on a single tenant is short; egress analysis and a browser-estate review add time, and a firm with two tenants or no existing logging takes longer. We set the timeline at scoping rather than quoting one in advance, because the honest answer depends on what your environment can report.
Will this disrupt our staff?
Usually not. The core review starts with the admin, network, browser and finance evidence you already have. If a surface is not currently logged or managed, we identify that at scoping and agree the next step before any endpoint change. We recommend telling staff the review is happening and why, because framing it as an investigation into misconduct tends to push the behaviour further out of sight.
What if we find client data went into a consumer AI account?
We document the technical facts available to support a breach assessment: what information can be confirmed, whose information, the service or account evidence, the exposure window, relevant permissions or vendor terms we can verify, and the evidence gaps. Whether that meets a reporting threshold under PIPEDA or a provincial regime is a determination for your privacy lead or counsel. We keep that line clear on purpose.
Do we need this if we already have an AI policy?
A policy written over an unknown estate governs very little. A policy written over an unknown estate leaves technical gaps. The review checks standing OAuth app permissions and other app access that a policy alone cannot show. Use the findings to decide which parts of the policy need technical enforcement.
How is this different from an AI readiness assessment?
A readiness assessment looks forward and asks whether the firm is ready to adopt AI deliberately. This review looks at what is already happening and what AI activity is already visible and where the evidence is incomplete. The two answer different questions. If the concern is unknown current AI use, start with discovery. If current use is already understood and the goal is planned adoption, start with readiness.
How often should we repeat it?
Treat it as a recurring control rather than a project. AI tools and staff both change, so a tenant reviewed once will drift. We recommend revisiting the discovery alongside your existing patching and access-review cadence rather than treating it as a one-off.
Start with the scoping call
Thirty minutes establishes what your tenant can report, whether a review is worth running, and what it would cost. Fusion Computing is CISSP-led. We have secured IT for Canadian firms since 2012 and now serve Toronto, Hamilton and Metro Vancouver.
Updated
