This blog was originally published by A1 Technologies here

Shadow AI Risk: How to Govern AI Use in Your Organisation

Your people are already using AI to get work done. They may be pasting information into public AI tools, using personal AI accounts for work, trialling new plugins or experimenting with AI agents outside approved systems.

This is shadow AI, the use of AI tools, applications or agents within an organisation without the knowledge, approval or governance of its IT or security teams.

For Australian organisations, the challenge is no longer simply whether employees are using AI. It is understanding which AI tools are being used, what organisational data they can access and what information employees are sharing with them.

The answer is not to block AI altogether. Organisations need a practical approach that protects sensitive information while giving employees secure, useful alternatives. For businesses already invested in Microsoft 365, this can include governed Microsoft 365 Copilot adoption alongside appropriate security, data protection and AI governance controls.

What is shadow AI?

Shadow AI occurs when employees use AI tools or services outside an organisation’s approved technology environment. It can include:

• Pasting sensitive company or client information into public AI tools
• Using personal AI accounts for work
• Installing unapproved AI plugins or browser extensions
• Connecting AI applications to corporate data
• Using AI meeting assistants without approval
• Trialling or deploying AI agents outside established governance controls

None of this necessarily happens with malicious intent. Employees are usually trying to work faster.

The problem is that IT and security teams may have little visibility over where organisational data is going, how it is being processed or whether the tools meet the organisation’s security, privacy and compliance requirements.

That creates risks around data leakage, inappropriate access, record keeping, inaccurate AI-generated information and the uncontrolled proliferation of AI applications.

Why does shadow AI create a data security risk?

AI does not create every data security problem from scratch. In many cases, it exposes or amplifies problems that already exist.

An employee who already has excessive access to information may be able to surface that information more easily through an AI tool. Sensitive information that has not been properly identified or protected may also be shared with an AI service without appropriate controls.

This is why AI governance cannot be separated from identity, access and data security.

Microsoft 365 Copilot, for example, operates within a user’s existing Microsoft 365 permissions. It can surface organisational information the user is already authorised to access. Poorly governed permissions and overshared information can therefore become an AI governance problem too.

Before organisations introduce more AI, they need to understand who can access their information, where sensitive data resides and what protections are already in place. Reviewing security and governance across your Microsoft 365 environment is an important part of establishing that foundation.

How do you identify shadow AI in your organisation?

The first step is visibility.

Organisations should establish what AI applications and agents employees are already using, which teams are using them and what types of organisational information are being shared.

This should include sanctioned AI systems as well as third-party and consumer AI services.

Technology can help identify AI activity and data risks, but it is only part of the process.

Organisations should also consider maintaining an AI systems register that records approved AI tools and material use cases, system owners, the types of data involved and relevant governance requirements.

This gives IT, security and governance teams a baseline for understanding where AI is operating across the business and assessing new tools as they are introduced.

You cannot effectively govern AI use you do not know exists.

What does shadow AI look like in Australian organisations?

Shadow AI can appear in almost any business function.

A professional services employee might paste client information into a public chatbot to help draft a document. A construction team might use generative AI to summarise contracts or safety procedures. HR or finance teams may experiment with AI tools that process employee or financial information. Someone may connect an AI meeting assistant to corporate meetings without considering where recordings or transcripts are stored.

The growth of AI agents introduces another consideration. Agents can interact with organisational data, applications and other systems, making visibility over what has been deployed, what it can access and who owns it increasingly important.

These behaviours are usually driven by genuine productivity benefits.

Simply banning AI can push that activity further underground. A more sustainable approach is to give employees clear guidance about approved tools, establish appropriate controls and provide secure alternatives that meet the same business need.

What are the key principles of AI governance?

AI governance can become complicated quickly. For most organisations, a useful starting point is much simpler:

1. Discover

Understand which AI tools and agents are being used across the organisation and what information they interact with.

Identify both sanctioned and unsanctioned AI use. Maintain an AI systems register covering material AI systems and use cases, establish ownership and understand the data involved.

Discovery should also extend to the data itself. Before deploying AI broadly, organisations need to know where sensitive information resides and whether employees have access to information they should not.

2. Govern

Set clear rules for acceptable AI use.

Employees should know which tools are approved, what information can and cannot be shared with AI systems, who can approve new tools and how exceptions are handled.

Governance also needs to extend beyond a written AI policy. Identity, permissions, device controls and data protections should support the rules the organisation has established.

3. Protect

Protect sensitive information at the data layer, not just at the AI application.

For organisations using Microsoft 365, this can include identifying and classifying sensitive information, applying sensitivity labels and using Data Loss Prevention policies to reduce inappropriate sharing.

These controls become increasingly important as organisations introduce Copilot and other AI tools that can interact with corporate information.

4. Monitor

AI governance is not a one-off project.

Monitor how employees use AI, review data security incidents and exceptions, and update policies as new tools and use cases appear.

Shadow AI can also tell you something useful. If employees repeatedly seek out a particular type of AI tool, there may be a legitimate business need that the organisation’s approved technology does not currently meet.

The objective should be to bring useful AI adoption into a governed environment, not simply block everything employees try.

From policy to practice: a 90-day shadow AI plan

You do not need a year-long program to start reducing shadow AI risk.

A practical 90-day approach can move an organisation from unknown AI usage towards more controlled and measurable adoption.

Days 1–30: Discover and establish policy

Start by identifying the AI tools and agents already being used across the organisation.

Document approved AI services, establish ownership and begin an AI systems register.

Create or review your AI acceptable-use policy. It should clearly explain which tools employees can use, what information should not be entered into public AI systems and how employees can request approval for new tools or use cases.

This is also the time to identify obvious high-risk use cases involving sensitive customer, employee, financial or organisational information.

Days 31–60: Understand and protect your data

Review the information your AI systems and users can access.

Identify your most sensitive information, review existing permissions and address obvious oversharing.

For Microsoft 365 environments, this may include implementing or refining sensitivity labels, information protection and Data Loss Prevention policies across collaboration workloads, endpoints and relevant AI interactions.

Where appropriate, start with user education and policy tips rather than immediately blocking activity. The objective is to change behaviour as well as enforce controls.

Days 61–90: Govern, pilot and monitor

Once the foundations are in place, expand access to approved AI tools through controlled pilots.

For organisations considering Microsoft 365 Copilot, start with groups that have appropriate permissions and data controls in place. Measure how Copilot is being used, identify issues and refine governance controls before scaling more broadly.

AI usage, data protection incidents, new AI systems and policy exceptions should then become part of the ongoing management of your Microsoft environment, rather than something reviewed only during the initial AI rollout.

The goal is not to eliminate every possible AI risk in 90 days. It is to establish an operating model that can evolve as AI use grows.

What should organisations know about Microsoft 365 Copilot governance?

Microsoft 365 Copilot operates within your existing Microsoft 365 security, identity and permission model.

That is important because Copilot does not automatically fix poor information governance.

If employees have access to information they should not have, AI can make that existing problem easier to surface. Before expanding Copilot, organisations should review identity controls, permissions, sensitive information and data sharing.

A secure rollout should therefore consider more than licensing and user training. It should address the data foundation underneath Copilot, including permissions, oversharing, information protection, Data Loss Prevention, auditing and ongoing governance.

How can Microsoft Purview help manage shadow AI?

Microsoft Purview can help organisations understand data risk and apply controls to sensitive information as AI adoption grows.

Its capabilities can support areas including data classification, sensitivity labels, Data Loss Prevention, auditing, insider risk, communication compliance and eDiscovery.

Purview can also help organisations gain greater visibility into AI activity and data exposure and apply controls to supported AI interactions.

The important distinction is that Purview is not an AI policy by itself.

Policy establishes how your organisation expects AI to be used. Technical controls help organisations discover activity, understand risk and put those policies into practice.

What are the limitations of AI governance controls?

No technology can completely eliminate shadow AI.

Data protection controls depend on organisations knowing where sensitive information resides and applying appropriate classification and policies. Legacy information may be poorly classified, permissions may be too broad and new AI applications can emerge faster than organisations can assess them.

AI agents add another layer of complexity because their access and behaviour depend on how they are built, what systems they connect to and what permissions they have been given.

This is why AI governance should be treated as an ongoing process rather than a one-time implementation.

Policies, technical controls and approved tools will all need to evolve as employees find new ways to use AI.

Building a governed approach to AI

Shadow AI is ultimately a symptom of something positive: employees see value in AI and are looking for ways to use it.

The challenge is turning that experimentation into governed adoption.

Start by understanding which AI tools and agents are already being used. Establish clear rules and ownership. Understand and protect the data underneath them. Give employees secure alternatives. Then continue monitoring and adapting as AI use changes.

A1 Technologies helps mid-market organisations across Australia and New Zealand secure and govern their Microsoft environments, including Microsoft 365, Copilot and the data and security controls that underpin AI adoption.

If shadow AI is already appearing in your organisation and you’re working out how to govern it without stopping useful adoption, talk to A1 Technologies about where to start.