From Shadow AI to Managed Governance: An MSP Playbook

From Shadow AI to Managed Governance: An MSP Playbook

AI is already inside customers’ environments. In many cases, it arrived before anyone decided who owns it, what it can access, or how it should be controlled. The central lesson for MSPs is clear: AI governance is an operational responsibility, not simply another product to resell.

It looks a lot like BYOD

The moment resembles the BYOD era, when executives brought in their own devices and expected IT to make them work. AI is following the same pattern: employees are already using it, often without approval. The biggest risk is shadow AI. Some companies assume no one uses AI because they have no AI policy, but people use it anyway. The gap is governance, not adoption—and that is where the MSP comes in.

The client owns the policy. The MSP makes it real.

The division of responsibility should be clear:

• The client owns and approves the AI policy. It sets their risk appetite. The MSP can help frame it, but shouldn't own it.
• The MSP sets up the infrastructure. That includes configuring the tool based on the policy.
• The MSP enforces the controls. That means access, connectors, and settings for data retention and privacy.
• The MSP produces the evidence. That means logs, approvals, offboarding records and a regular report the client can review and sign.

Most organizations stop at a written policy, but that is not enough. Typing “please don’t hallucinate” or “don’t send confidential data” into an AI tool is a request, not a control. A real control is a setting that can be enforced and demonstrated.

Connectors and data access need attention

Connectors are an easy risk to miss. When an AI tool connects to cloud storage, email, or a calendar, it can inherit the user’s access. If an administrator connects a company-wide drive, the AI may gain access to the entire repository. Start small with read-only access, then expand step by step. Each change should be reflected in the policy, settings, and evidence.

A recurring service, not a one-time project

The setup is a project. The ongoing work is what turns it into a recurring service:

• Quarterly reviews of policies and controls
• Access changes as more users and connectors come online
• Monthly evidence reports
• Training and change management
• Deploying and reviewing custom automations

This service may not appear as a separate line item. It can be positioned as an expansion of the managed services already being provided.

Fundamentals first

AI will not fix a poorly run environment. It amplifies what is already there, good or bad. Clients that have not adopted MFA, password management, or sound change management may still ask for AI, effectively skipping the basics. Start small, prove value, and then expand. Clients will pay for disciplined guidance—even when the right answer is no.

Treat AI like a service account

For now, treat AI like a service account or a new hire with a user license. It needs tracking, oversight, human review, and evidence. MSPs already know how to manage those responsibilities.

The takeaway

When a client asks about AI, begin with a review of its policy, risk tolerance, and objectives. If the required expertise is not available in-house, bring in a qualified partner. A trusted advisor does not need to perform every specialized task, but should coordinate the right experts. Selling a Copilot license is only the beginning. The MSPs that help clients govern AI—from policy through controls to evidence—will be the ones clients rely on.