THE QUESTION THIS NOTE ANSWERS
What is shadow AI?Shadow AI means AI use outside an organization’s approved governance. An application name or permission grant alone does not establish that use is unapproved. Establish the evidence, then confirm purpose, ownership and approval.
What we read / what we never read
Public-domain evidence and customer-approved, read-only tenant evidence have different boundaries. No message bodies, files, chats, documents or prompts. Public availability is not blanket legal clearance; confirm scope and obtain qualified advice.
EXPLANATORY DIAGRAM · EVIDENCE AND LIMITS
Connected app → granted access → evidence to investigate.
Inventory
Identify the app, publisher and owner from supported sources. Presence alone is not misuse.
Access type
Separate delegated permissions from application permissions; review consent and resource scope.
Credentials
Review supported credential metadata and expiry. A user password reset does not revoke every app credential.
Decision
Confirm business purpose, activity evidence and the owner before agreeing a change.
Shadow AI means AI use outside an organization’s approved governance. An application name or permission grant alone does not establish that use is unapproved. Establish the evidence, then confirm purpose, ownership and approval.
Shadow AI is an ownership question before it is a tool question.
Short answer: Shadow AI means AI use outside an organization’s approved governance. An application name or permission grant alone does not establish that use is unapproved. Establish the evidence, then confirm purpose, ownership and approval.
How shadow AI gets in
Nobody installs shadow AI. An employee signs in to a new assistant with their work account, sees a screen asking for access to mail and files, and clicks yes because that is what the screen is for. The tool now holds a token with whatever it asked for. Multiply that by every curious person in a thirty-seat office and you have a tenant full of connections the owner has never heard of.
Kaseya's 2026 SaaS Security Report, built from 27.6 billion security events across more than 50,000 small-business environments in 2025, names this as one of its six major findings: AI assistants, automation tools and collaboration apps are spreading through business tenants on single sign-in grants, and each one is another connection into email, files, calendars and chat.
The report makes a point every MSP should repeat to clients: a grant of this kind can stay active through its token after the user's password is changed or reset. Resetting the password after a scare does not close the door the app came through.
Why MSPs keep underselling it
Clients want this conversation. In Kaseya's 2026 State of the MSP Report, a survey of 1,061 MSPs, 48 percent said AI and automation will be their clients' top IT need this year, ahead of security at 42 percent. Yet only 13 percent of those MSPs said AI and automation is a meaningful revenue source today.
That gap is a packaging problem. Most MSPs are waiting for a client to ask for "AI services." The owner is not going to ask for a product category. The owner will react to a sentence like "four AI tools have standing access to your finance manager's mailbox, and two of them were approved by people who no longer work here."
The same survey found 34 percent of MSPs expect AI to increase security risk for their business in 2026. Shadow AI is where that risk shows up first in a small tenant, and it is the easiest AI conversation to start because it begins with something the client already did.
What to look for in a tenant
Four questions cover most of it. Which third-party apps hold consented access, and to what (mail, files, calendar, directory). Who granted each one, and is that person still employed. When each app last signed in. Whether any grant is broader than the tool plausibly needs, such as a note-taking assistant with read access to every mailbox.
Kaseya's recommendations line up with that list: audit connected apps and remove the unused ones, hold third-party permissions to least privilege, require review before new connections are approved, and watch for unusual consent requests.
None of this requires opening a single email or document. The grants, the scopes and the sign-in dates are configuration. That matters when you are asking a prospect for access before they are a client.
Turning the finding into a service
Once the list exists, the service writes itself. A one-time cleanup: revoke stale and over-scoped grants, set consent so new connections need approval. A policy: which AI tools the company endorses and what data they may touch. A monthly review: new grants since last month, reported to the owner in plain language. That is a defined AI governance offering with a scope and a price, which is exactly the packaging the 13 percent revenue figure says most MSPs have not built yet.
Frequently asked questions
What is shadow AI?
Shadow AI means AI use outside an organization’s approved governance. An application name or permission grant alone does not establish that use is unapproved. Establish the evidence, then confirm purpose, ownership and approval.
Is ChatGPT automatically shadow AI?
No. Classification depends on actual use and the organization’s approval and governance—not the application name.
What are examples of shadow AI?
An AI meeting assistant that joins calls and reads calendars, a note-taker with read access to every mailbox, a browser extension summarizing email, an automation tool connected to files and chat, and AI features switched on inside existing SaaS tools during an update. Kaseya's 2026 SaaS Security Report groups these under connected apps spreading on single sign-in grants.
What is the difference between shadow AI and shadow IT?
Shadow IT is any software or service employees use without IT approval. Shadow AI is the subset that uses AI and, more often than older shadow IT, holds a standing permission into email, files and calendars. The risk is the same category, unreviewed access, with a larger blast radius because the tool was granted access to content rather than installed on one machine.
Why is shadow AI a security risk?
Connected AI apps can hold standing access to business data through tokens that stay valid after a password change, according to Kaseya's 2026 SaaS Security Report. If an app is malicious, compromised or simply over-permissioned, that access is invisible to anyone not reviewing the tenant's app grants.
How common is AI demand among MSP clients?
In Kaseya's 2026 State of the MSP Report, 48 percent of 1,061 MSPs named AI and automation as their clients' top IT need for 2026, yet only 13 percent reported AI as a meaningful revenue source.
Does a permission grant prove active use?
No. It establishes supported access evidence. Activity requires an appropriate source and time window.
How should an MSP start an AI-governance conversation?
Bring a dated inventory, supported permissions and explicit unknowns. Ask the customer to confirm ownership, purpose and approval before recommending work.
Can SCOUTz discover all Shadow IT or embedded AI agents?
No. Supported tenant inventory is one evidence layer, not complete endpoint, network or runtime coverage.
How SCOUTz gets you there
Finding an AI application is not the same as finding unapproved AI use. SCOUTz reviews supported application inventory, ownership, permissions, consent and credential metadata through customer-approved, read-only tenant access. Where an approved source supplies activity evidence, show its coverage and date. Otherwise actual use is not assessed. The customer confirms purpose and approval. This is not complete Shadow IT discovery, AI-agent runtime inspection or access to private prompts and content.