Strategy
Closing the AI governance gap

Across our AI discovery assessments, one finding stands out every time: a clear gap between what leadership believes is happening with AI and what the usage data actually shows.
Most AI risk sits in that gap. So that’s where good AI governance starts. You can’t manage what you can’t see.
01 · The pace problem
Why AI governance falls behind AI adoption
AI use is growing faster than most organisations update their rules. Policies are usually reviewed every year or two. Staff can start using a new AI tool in an afternoon.
≈2× In one recent assessment, use of the approved AI tool roughly doubled in just three months. A policy written before then no longer described how people worked.
This isn’t a failure by the people who write policy. It’s a speed problem. That’s why AI risk management needs real evidence, not just documents.
02 · Shadow AI
Shadow AI: on paper versus in practice
Here’s a pattern our assessments consistently uncover. The organisation runs Microsoft 365. It has strong security, clear policies and one approved AI tool. On paper, everything looks under control.
Then we look at the usage data: the system records that show which apps people actually use and what those apps can reach.
On paper
One approved AI tool for the business
A clear policy on which AI apps are allowed
Microsoft 365 protected by mature security
In practice · the usage data
Public AI tools used almost as widely as the approved one
Dozens of generative AI apps in use, none formally assessed
Outside apps that can read email, files or meeting transcripts, with no record of anyone approving them
This is shadow AI: AI tools people use at work without formal approval. It’s rarely malicious. People simply want to work faster.
None of this means AI should be blocked. Many quick wins are simple IT fixes. Control which apps staff can connect to company data. Review what those apps can access. Sort the apps already in use into approved, restricted or blocked. Shadow AI detection starts the conversation. It doesn’t end it.
03 · AI privacy and vendors
Third-party AI apps and your Microsoft 365 data
The harder questions are for the companies behind these apps. When an AI app connects to Microsoft 365, we ask three things:
Where does the data go? Does the app read your data where it sits, or copy it to its own systems?
Who else is on those systems? Is your data kept separately, or on a service shared with other customers?
What happens to it later? How long is it kept? Who can see it? Is it used to train AI models? How is it deleted?
A connection to Microsoft 365 doesn’t mean the data stays in Microsoft 365.
The question behind every AI privacy review
This is where AI security and AI privacy meet. An app can look harmless on the permissions screen and still move your data elsewhere. The vendor’s answers belong in your AI risk assessment, not just the purchasing file.
04 · Copilot governance
Microsoft Copilot governance: the approved tool needs governing too
It’s easy to assume the approved tool is the safe one. But approval is the start, not the finish. In our Microsoft 365 Copilot security reviews, we find the same three gaps again and again:
Gap A · Data leak controls
Rules that stop sensitive data leaving the business (often called DLP) don’t yet cover what people type into AI or what AI produces.
Gap B · Sensitivity labels
Confidential files aren’t tagged as confidential. So the AI can’t tell what should stay restricted.
Gap C · Staff-built agents
Staff build their own AI assistants. Often no one has checked what those assistants can access.
A good Copilot readiness assessment checks all three before use grows, not after. This is the practical side of AI data loss prevention.
05 · Discovery first
Why AI discovery comes before any AI governance framework
This is why AI discovery matters. The goal isn’t a list of apps to ban. It’s a clear view of how AI is really used, where data goes and where controls fall short. With that view, you can decide what to block, what to protect and what to enable.
An AI governance framework built on evidence works better than one built on assumptions. Your AI acceptable use policy and generative AI policy can then match how people actually work. That makes them easier to follow. This is responsible use of AI in practice: clear rules, based on reality. It’s also the foundation of our AI strategy and advisory work.
06 · Practical steps
What to do about it
Check real use, not just the policy
Find out which AI tools people actually use, how often and in which teams. Include public tools and AI features built into other apps.Review every connected app
List the outside apps that can reach email, files or meeting transcripts. Ask each vendor the three questions above and record the answers.Govern the approved tool properly
Close the common gaps: data leak controls, sensitivity labels and checks on staff-built agents.Decide, then keep it current
Choose what to block, protect and enable. Then review often enough to keep up with how fast use is changing.
RegReady AI · AI governance Australia
Where this meets the 10 December privacy rules
Australian organisations have one more reason to know what their systems really do. From 10 December 2026, privacy policies must explain when software uses personal information to make, or help make, decisions that significantly affect people (APP 1.7–1.9). You can only explain what you’ve found.
RegReady AI is our service for exactly this. We find the automated decisions in your systems and record each one in a register. We then draft plain-English wording for your privacy policy, which a legal partner signs off. Choose a 4–6 week Sprint, a 2–3 week Fix to close specific gaps, or a 12-month Watch to keep the register up to date. Read our guide to Australia’s AI privacy rules from 10 December 2026.
General information only, not legal advice.
FAQ
Frequently asked questions
What is shadow AI?
Shadow AI is any AI tool used at work without formal approval. It includes public chatbots, AI features inside other software, and outside apps that staff connect to Microsoft 365.
Should we block public AI tools?
Not by default. Blocking often pushes use out of sight. It’s better to see what’s actually in use first. Then decide what to block, what to protect and what to enable.
We already have Copilot and a security team. Isn’t that enough?
It’s a strong start, but the approved tool still needs governing. We often find gaps in data leak controls, sensitivity labels and staff-built agents, even in well-run Microsoft 365 environments.
How does AI discovery relate to the 10 December 2026 privacy rules?
The new rules (APP 1.7–1.9) require privacy policies to explain when software uses personal information to make, or help make, significant decisions. You need to know where those decisions happen before you can explain them. This is general information, not legal advice.
The goal isn’t to restrict AI. It’s to build the confidence to use it responsibly. So, how big is the gap between your AI policy and what your people really do? If you’re not sure, that’s the place to start.
Navigator AI Advisory
See what’s really happening with AI in your organisation.
In 30 minutes we can work out where your AI use stands and what to do first.
Book a 30-minute consultation →
Related services
This article is general information and isn’t legal advice. Navigator AI Advisory is part of AI4 Group.















