Skip to the main content.

Support My IT

Create a secure, scalable technology foundation that keeps your business productive and growing.

Untitled design (3)

Secure My Business

Reduce risk, strengthen resilience, and stay ahead of evolving cyber threats. 

Untitled design (3)

Automate My Business

Eliminate manual work, increase efficiency, and empower teams with AI and automation.

Untitled design (3)

Modernize My Technology

Transform legacy systems into modern platforms that improve performance and accelerate innovation.

Untitled design (3)

Commercial Industries

We understand what most managed service providers don’t – when it comes to industry-specific technology, one-size-fits-all solutions don’t exist.

Untitled design (3)

Public Sector

Specialized IT, cybersecurity, and compliance support for schools, BOCES, local governments, and utilities — backed by 40+ years of public-sector experience.  

Untitled design (3)

Locations

We serve clients across the lower 48, with regional concentration in the Northeast, Mid-Atlantic, Southeast, Mountain West, and West.  

Untitled design (3)

The Sourcepass Story

Built and run by technology, security, and managed services people who were tired of how IT gets delivered – and decided to do it differently. 

Untitled design (3)

The Sourcepass Experience

Excellent service, strategic guidance, and technology delivered with innovation – the operating model behind every Sourcepass engagement, across IT, security, Microsoft, and AI. 

Untitled design (3)

 

AI Governance for SMBs: Microsoft 365 & Copilot Guide

 
AI Governance for SMBs: Microsoft 365 & Copilot Guide

AI is no longer a side experiment. It is part of everyday work. Employees use Microsoft Copilot, generative AI apps, browser tools, and other platforms to draft emails, summarize documents, analyze data, write code, and automate routine tasks.

For SMB leaders, the question is no longer whether employees should use AI. It is how to set up AI governance that supports adoption. That means protecting sensitive data, keeping people accountable, and managing business risk.

In Microsoft 365, AI governance is tied closely to identity security, permissions, information governance, and employee behavior. Microsoft Copilot follows your existing Microsoft 365 permissions. Users can surface any information they already have access to. Microsoft's Copilot privacy and security guidance makes one point clear. AI security depends on the access controls already protecting your data. 

An AI governance program helps SMBs decide which tools and use cases and what data AI can access. It also defines when people must review AI output and how to tell if controls are working. 

 

What is AI Governance?

 

AI governance is the set of policies, roles, technical controls, review, and monitoring a company uses to manage AI.

It should answer a few questions:

  • Which AI tools are approved for business use?
  • What data can employees enter into those tools?
  • Which business processes can use AI output?
  • Which tasks need human review or approval?
  • Who can approve new AI apps and integrations?
  • How are outside AI vendors evaluated?
  • What happens when an AI security or data incident occurs?
  • How will leadership know if controls are working?

The goal is not to write a rule for every AI scenario. It is to give employees and leaders a repeatable way to evaluate new uses as they come up.

The NIST AI Risk Management Framework is a useful starting point. NIST describes it as a voluntary framework for building trust into how AI is designed, used, and evaluated. NIST also offers a Generative AI Profile that covers risks specific to generative AI. 

For an SMB, this does not require a large new team. Existing IT, security, legal, HR, compliance, and operations roles can share ownership based on company needs. 

 

Why SMBs Need AI Governance

 

Governance problems often start before a company formally rolls out AI. 

Employees may already use AI on their own to research topics, draft documents, summarize information, review code, or analyze data. This creates a mix of approved and unapproved AI use that is hard for IT and leadership to track.

Banning AI is not the answer. Employees need clear limits on what tools they can use, what data they can share, and when they need approval.

A useful AI acceptable use policy should cover these areas:

  • Approved AI platforms and accounts
  • Acceptable business use cases
  • Restricted and prohibited data
  • Human review requirements
  • Third-party AI apps
  • Browser extensions and AI integrations
  • App and data access
  • Vendor approval
  • Incident reporting
  • Ownership and accountability

This approach matches Sourcepass' internal AI governance model. Our  Generative AI Usage Policy organizes AI use around approved platforms, data classification, human oversight, least privilege access, training, vendor risk, auditing, incident reporting, and accountability.

 

Microsoft Copilot Governance Starts with Identity and Data

 

In Microsoft 365, Microsoft Copilot governance should not sit as a separate layer on top of your security setup. It should build on it. 

Copilot reaches company data through Microsoft Graph. It can use documents, email, calendars, chats, and meetings that a user can already open. Microsoft states that Copilot only surfaces data for which the user has at least view permission to see. 

That makes your current permissions very important.

If employees have more access than they need, AI makes that problem easier to see. Users can now ask questions across everything they can reach. 

 

Review Access Before Expanding AI Use

 

Before a broad Copilot rollout, SMBs should review these areas:

  • Microsoft Entra identities and privileged access
  • SharePoint and OneDrive permissions
  • Group memberships
  • External sharing
  • Site and content ownership
  • Data classification
  • Retention policies
  • App and agent access
  • Connected data sources

Microsoft recommends using the permission models in services like SharePoint so the right people have access to the right content. Copilot respects those permissions.

This changes the question from "Can Copilot see our data?" to a better one. Does each employee already have the right access to the data Copilot can help them find?

You can measure that and act on it. 

 

Create an AI Acceptable Use Policy Employees Can Follow

 

A policy only reduces risk if employees can use it in real work.

Skip vague guidance like "use AI responsibly." Instead, sort AI activity into approved, restricted, and prohibited uses.

 

Approved Uses

 

Depending on the company and the data involved, approved uses could include the following:

  • Drafting internal communications
  • Summarizing approved information that is not sensitive
  • Brainstorming ideas
  • Building document outlines
  • Improving existing documentation
  • Supporting research that will be checked independtly

 

Restricted Uses

 

Some tasks should trigger extra review because the data or decisions carry more weight. These could include the following:

  • Customer or employee data
  • Confidential contracts
  • Financial data
  • Regulated records
  • Proprietary source code
  • Legal or compliance work
  • AI integrations with business systems
  • Automated decisions that affect employees or customers

Base restrictions on your own data classifications, contracts, regulations, and risk tolerance. A generic list will not fit every company. 

 

Prohibited Uses

 

Clearly define when AI use creates too much risk. This includes legal, privacy, security, and ethical risk.

Your AI acceptable use policy should make these limits easy to follow. Employees should know what they can do without approval, what needs review, and who to ask when they are unsure.

 

Require Human Review Where Decisions Matter

 

Fast AI output is not automatically correct. 

Microsoft notes that generative AI responses are not always accurate. It advises users to review AI output before sharing it. 

Build human review into the workflow. Do not treat it as an optional last step. 

Define which outputs need review for these factors: 

  • Accuracy
  • Confidentiality
  • Proper use of source material
  • Bias or flawed conclusions
  • Regulatory impact
  • Business impact

The person who approves the work stays accountable for the result.

For higher impact processes, go further. You may require documented approval, named sources, tighter access, or expert review before AI recommendations are used.

 

Evaluate Third-Party AI Vendors Before Sharing Business Data

 

Copilot is only one part of the picture.

AI features now appear in business apps, browser extensions, meeting tools, development platforms, analytics products, and tools that departments buy on their own.

Before approving an outside AI service, ask these questions:

  • What company data does it collect?
  • Where is that data stored?
  • How is access controlled?
  • What retention options exist?
  • How are prompts and AI interactions handled?
  • Can your data be used to train models?
  • What audit logs are available?
  • How does the vendor report security incidents?
  • Can your data be deleted or exported?
  • What integrations or permissions does it need?

Check the answers against the vendor's actual contracts and technical documents. Do not assume one AI service works like another.

Employees should also know that business AI accounts and personal AI accounts often have very different security and data controls.

For Copilot, Microsoft states that prompts, responses, and data accessed through Microsoft Graph are not used to train its foundation models. Microsoft also provides admin controls for manage Copilot access and features. 

 

Connect Responsible AI to Microsoft 365 Security Controls

 

AI governance works best when policy is backed by technical controls.

Depending on your licensing and setup, these controls may including the following: 

  • Microsoft Entra identity and access controls
  • SharePoint and OneDrive permissions
  • Microsoft Purview sensitivity labels
  • Microsoft Purview retention policies
  • Data Loss Prevention
  • App and agent governance
  • Microsoft 365 admin controls
  • Audit and compliance tools

Microsoft states that Copilot respects usage rights on data protected by Microsoft Purview, including sensitivity labels. Admins can also manage stored Copilot interaction data through Purview. 

The idea is simple. Policy sets what employees should do. Technology helps enforce it. 

 

Build a Risk Based AI Use Case Review Process

 

Not every AI project needs the same level of review.

A simple internal drafting task should not get the same scrutiny as an AI system that affects hiring, finances, healthcare, legal advice, security actions, or customer access.

A short AI intake form can capture the basics:

  • Business purpose
  • AI platform or vendor
  • Users and departments involved
  • Data the system will access
  • Expected output
  • Decisions the output will influence
  • Existing security controls
  • Required human reviewer
  • Integration or app permissions
  • How success will be measured
  • Known risks or exceptions

From there, route each request based on its risk. 

Match the review to the risk. Spend the most effort on use cases that involve sensitive data, heavy automation, broad permissions, or decisions where a wrong answer is costly. 

 

Measure AI Governance Through Outcomes, Not Usage Alone

 

Adoption numbers show whether people use AI. They do not show whether they use it safely or well.

Strong AI governance measures both adoption quality and risk.

Useful measures include the following:

  • Shared of targeted employees who finished AI training
  • Number of approved AI use cases
  • Number and type of policy exceptions
  • Completed human reviews
  • Share of AI vendors that passed security review
  • AI incidents involving sensitive data
  • Common causes of AI incidents
  • Time to resolve reported AI concerns
  • Permissions problems fixed through governance reviews

Pick metrics that match your own risks and controls.

If employees do not understand the policy exceptions. That tells you more than raw usage. If overshared Microsoft 365 data is the concern, track permission fixes and access review results. 

 

Treat AI Governance as an Ongoing Security Process

 

An AI system's risk can change after launch.

A new connector can expose more data. A permission change can open access to new files. An agent can gain new abilities. A vendor can change its data terms. A team can start using an approved tool for a task no one reviewed. That is why governance needs regular reviews. 

Review these areas on a set schedule: 

  • AI tools and integrations in use
  • Accounts and permissions
  • Connected data sources
  • New business use cases
  • Policy exceptions
  • Vendor changes
  • Security incidents
  • Employee feedback
  • Legal and regulatory requirements

Microsoft's Copilot management framework follows a similar model. It covers security and governance, management controls, and measurement and reporting, including lifecycle management for agents and connectors. 

Quarterly reviews are a good starting point for many SMBs. Add extra reviews after major changes in technology, regulations, security, or business operations. Adjust the schedule to fit your risk profile. 

 

Include AI in Incident Response and Security Awareness

 

Employees need to know what to do when AI use goes wrong. 

Add AI scenarios to your existing incident response plan. Examples include an employee entering confidential data into an unapproved tool, a compromised AI account, or AI output that causes a serious business error.

Your response plan should name who receives the report, how to assess the affected data, when to restrict access, what logs to keep, and how to document fixes.

AI should also be part of security awareness training.

Instead of telling employees to "be careful with AI," teach them to stop and ask a few questions: 

  • Is this AI tool approved?
  • What type of data am I using?
  • Does the tool need this data to finish the task?
  • Does the output need to be checked?
  • Is this a high impact decision that needs expert review?
  • Do I need approval before connecting this tool to company data?

This is where policy becomes habit. Employees should be able to turn governance rules into quick, consistent decisions during normal work.

 

Building Responsible AI Governance for Microsoft 365

 

Good AI governance does not mean predicting every way employees will use AI.

It means clear ownership, practical limits, the right technical controls, and a repeatable way to decide when extra review is needed.

In Microsoft 365, the foundation is the environment AI already relies on. That includes identity, access, permissions, information governance, and employee behavior.

Sourcepass' follows a similar model in its own governance materials. It connects responsible AI adoption with approved tool lists, vendor reviews, Conditional Access, Data Loss Prevention, employee training, AI approvals, risk classification, logging, and shadow AI discovery. SMBs can adapt this framework to their size, regulations, and risk profile. 

The goals is not to block AI or to adopt it everywhere. It is controlled adoption. Leaders should know which AI tools are in use, what data they can reach, who owns the output, and whether controls are working. 

FAQ