Defend, Govern & Prove:
Managed AI and data compliance

Welcome to the episode seven really of managed AI and data security on our defend govern journey. This episode is probably one of the most important ones we kind of talk about now specifically in kind of 2026 and moving forward. We've had AI now and this is being recorded on in February. It's around about 43 44 months. I think AI has been misconceived sometimes. We think of it and it's all actually automations and automation scripts we're building.

This one here is really designed for us to kind of identify what do we mean by the AI gold rush because it's essentially here we've got Copilot and word ChatGPT in browsers. We've also been talking about AI productivity for god knows how long. Copilot really starts to write our emails, summarize meetings, building proposals. But I also find that very more often than not at organizations kind of tend to take a stop and forget to ask a few questions or stop to think about questions that we should be asking like what data is AI actually allowed to see as an example.

It exposes the risks that already sitting quietly inside our tenant. And this is why this topic is so important. Today isn't really here to talk about anti-AI session. This is about running AI safely in a scale and at scale, sorry, and being able to prove that you did. And realistically, we're in the same phase cloud was in and around 2020 in 2016. Everyone wants the benefits first, governance later. Marketer teams are using image generators, sales teams paste contracts into ChatGPT.

We have so many risks when we talk about AI at the moment. And there's very little change in posture that we should we are or should be adopting. And this conversation and topic really is about where do we go? How do we take this journey? And what's next on this platform? So let's look at the real threat. Poor data management. Copilot doesn't make tenants unsafe. And I think this is so relevant to today's topics. Everyone thinks that this is a breach, it's a concern, it's a risk.

The reality is that's not the case. We actually can see the information that we're looking at and all Copilot's doing is just bringing it to the forefront of the conversations we're having. Copilot doesn't make our tenant unsafe.

Copilot only reads what users already have access to and it explain in kind of explanational terms or providing reason behind it sits firstly on Microsoft Graph. It leverages Work IQ to build that context and delivery. It looks at the permissions that we currently have around SharePoint, OneDrive, and Teams. So, if permissions are messy, AI just operationalizes that mess. We start to see the messy environment that we've already got configured. It's going to bring it to the forefront.

If everyone can read HR, your AI assistant can essentially summarize the HR files that sit there. That's how risky it can be if we haven't got the right structures in place.

The other elements I think we're already forgetting and actually well we're not forgetting we already know they exist and they're areas we need to focus on is shadow AI the hidden epidemic and it is very much an epidemic now when you think about this there are unsanctioned tools in the environment that we're not managing AI ChatGPT custom APIs to business data without IT approval or oversight all this information is exposure and we're running too many risks the ability for our HR people to get documentation summarized because ChatGPT is convenient and they're uploading contractual information.

If it's the free version, that model can be trained on. They can start training on this information, feeding it back to the big LLM. We also have compliance blind spots when we're not controlling it in an internal environment, you know, utilizing data loss prevention, Purview labeling, controls within SharePoint, our conditional access policies to restrict data coming in and out. We start having compliance blind spots, the ability for people to be able to do things without us knowing.

Dropbox accounts being created. If you look at this whole broad picture, we have shadow it that we need to control. That's things like your Dropbox, your Gmail box, your what's it called? Your Google Drive, sorry, and box that we need to manage and maintain. That's how shadow AI shadow it, sorry. Shadow AI though is much worse. It's much harder for us to understand. And it's much harder for us to dictate. Users don't download data anymore.

What they start doing is they're now pasting it into a product. They're pasting it into AI. And one, once that text leaves our tenant, your compliance at this point tooling has never seen that information being exposed. When we copy and paste without the right controls, we're not auditing what's taking place. There's no audit trail. There's no data loss prevention policy configuration. And there's no retention labeling, no legal hold on those applications, which means it's free for people to use.

Friday afternoon installs is another top topic hot topic when we talk about this. Believe it or not, it happens a lot when we're winding down for the end of the week. People quite often go, "Well, I've got these last bits. I wonder if I got a couple of hours. Let me just play around with these products. See if I can get some information that's useful." 27 unofficial AI tools installed in minutes that it discovers months later after a breach.

It's happening today. It's not just AI. It's any kind of specific product around shadow IT. On a Friday, we quite often are winding down. We leave it. Saturday, Sunday goes by. We're back into Monday. We're back into the full swing of things. These things get missed. That's why I'm classing it as a hidden epidemic. We don't always know how do we build the right safety and security. And this is where Microsoft comes in. We talk about Microsoft Copilot.

It's our top topic this year for inforcer. Built-in AI safety architecture is absolutely pivotal. The ability for us to be able to create those controls.

And the good news is Microsoft has actually started to prepare for this. It's taken them a while, but they now have some controls and configurations in place that we can start to leverage to better this product delivery. Key statements that I want to identify here. AI does not bypass security. AI amplifies security. So if good governance becomes good governance as a result becomes safer, bad government governance starts to become a catastrophe ability where it all starts to go wrong as a result.

And these are the five layers I really want you guys to take home. The ability to kind of think outside the box. But these classification layers are absolutely crucial when we talk about the right governance for AI within the Microsoft 365 environment. Data classification Purview sensitivity labeling is one of the hot topics that are recommended by Microsoft to have deployed. We know fundamentally it's a difficult thing to do because most Purview data classification delivery actually requires significant training for users.

It's time consuming. It's not something we always think about but it's absolutely should be on the forefront of conversations we have with customers. The other one's data loss prevention. Really simple. Now it actively block sensitive information being copied into AI tools. We can actively manage that. If we're using edge and we're using mcast or edge browsing, which is now available for edge mam, we can start to prevent this information from going to the wrong controls. But we can also leverage data loss prevention to immediately stop sensitive information going to Copilot.

That's now available since Microsoft Ignite that has been pushed out and it's available for us to use. The other area and probably the most important one is access controls. Conditional access and enter ID for us can really start to leverage and restrict the right controls around how we access and manage data. So requiring device compliance in order to access our information. Huge hot topic, hugely deliverable and hugely valuable. We then leverage sorry defender for endpoint the ability to start making additional controls and modules to prevent information from being extracted.

Using edge browser management we can also restrict what information is copied and pasted plus monitor it. Defender for cloud apps allows us to start sanctioning and unsanctioning applications if we decide to go to that P P2 plan. Audit and logging clearly the most important topic when we talk about governance and proof. We need to be able to audit and track every prompt request and data movement across AI services that we leverage and we have to be able to do that.

If we are working for vertical industries that require frameworks and regulations, this is a requirement we have to consider. And then the last topic which is nice and simple is governance based governance dashboards. That single pane of glass the visibility for you guys to see Purview DSPM for AI can deliver that element for you to be able to see what needs to go on. For us visibility is so important. We need to be operationally aware not policy screenshots.

We need to be able to see that continuous ability and visibility as we go through. So what do we do when we talk about detecting shadow AI? The new insider threat is essentially the one area that we need to as MSPs and as customers consider and think about. It works with detection methods.

So defender for cloud apps discovery which is part of business premium. If you're using defender for endpoint we can see what's going on. We need to know the controls the measures. What are we installing on our devices? What are we searching through browsers? Well cloud app discovery allows us to see that. Two different types. Cloud app cloud defender for cloud apps and cloud apps discovery. They're all part of the security center. They're all available in cloud app discovery.

In cloud apps and you can see the discovery element Purview audit tracking. Another hot topic of requirements. The ability to see data egress taking place. We can start controlling and measuring that or measuring it to start with business premium and taking more of a control approach controlled approach sorry when we want to start monitoring and detecting risks. And the other area is conditional access. And I go back to this. Everything we talk about when we deliver these sort of elements, it's identity based.

Everything we access in a Microsoft 360 36 365 environment starts with identity. Unless we're using API ingestion and enterprise apps, they're not built without identity as in the first place. Everything starts with identity. When we talk about these threats, threat landscapes and the ability to manage this, we have to be able to detect this and we have to be able to manage it. And sometimes it comes down to us to create those responses as a result.

So one of the things I always want people to take away with is acting before shadow risk becomes a shadow breach. We don't always see it. We don't always have the ability to understand what's going on or how do we deal with these indications and vulnerabilities, but we need to. If you can detect shadow AI after a breach, you're not doing security. What you end up doing is incident explanate explanation. We need to be at the forefront of reducing these risks, breaching that gap, preventing those gaps from being a problem.

And that starts with making sure we look at a zero trust model when we start to build out this best practice delivery. So how do we go about this and leverage or utilize inforcer really as one of those pillars that we can implement the right controls and that's AI readiness and proof of governance.

So one of the benefits of utilizing inforcer at the moment is we can bring together a Copilot readiness assessment, the ability to see the controls that are recommended. Do we have Purview in place? Do we have the right conditional access data loss prevention? Are our permissions tidied up enough? We can start seeing that single pane at the very bottom of one of those reports that shows you your SharePoint overview. This information is invaluable when we start talking about an AI journey with users.

Whether that's with Copilot or with another AI product like ChatGPT enterprise. This report is significantly valuable because it talks about the security foundations, the data governance, our technical readiness. It's the conversations we have to have with customers before they even consider using AI properly. The other area is shadow AI detection. So although we can't configure and see this directly within a customer's environment, it's 100% available for us to look at within MCCAST within shadow app discovery, cloud app discovery.

And we have loads of information on this. You can log into the security center, drop down cloud apps and then select discovery, see what's going on in the environment. This only works if we deploy defender for endpoint to intune based managed devices or two devices that we manage for that customer. So it's absolutely available and we reason why I've added it in here or the scripts have added this in is specifically because we can deploy those policies which then means we can then measure against our customers.

And the last one is framework mapping the ability to start checking that where customers align to the right frameworks for these controls when it comes to readiness. All of this is 100% available for us to manage, monitor and implement and we can utilize the inforcer platform to start that journey with our customers.

So areas of governance, one of the areas, probably the most important area when we talk about best practice and starting that AI journey is we have to continue the governance and it has to be at scale as MSPs. So we start with drift detection. We need to know what policies are being changed and when they're being changed because the fundamental truth is we need to be able to identify those drifts and remediate in real time those concerns that happen and run.

The ROI dashboards return on investment for me is critical. Entry ID dashboard that we have based in inforcer is your single point of a hot topic for customers. When we talk about permission management and readiness that starts and it stems with tidy Microsoft 365 environments and a tidy environment starts with identity. So we have the ability to see the entry ID dashboard measure information identify risks and address somewhere where required. And the last area is multi-tenant view, monitoring your AI governance.

And for you guys, you can do that within inforcer across the board. And we talk about best practice standardization, the alignments of your right security controls. So we should be implementing the right controls for the industry verticals we work in. And we can see that as a measurement in alignment, the ability to really start to clean up those controls, measurements, and delivery. inforcer makes you the AI governance partner your clients didn't even know they needed.

So we can start that new journey with customers as we go. So let's talk about key takeaways. AI brings productivity and risk in equal measures. And it's so important to know that if you haven't got a clean environment, you're not going to get a clean output from Copilot. Every AI tool accelerates work and it expands the your attack surface without proper governance. We have to be at the forefront of delivering this successfully. Microsoft security stack is AI ready.

It is. We have purview and defender controls we can absolutely implement and get controlled and configured and running immediately. Shadow AI is a hidden compliance threat as a result of this. Unsanctioned tools start to bypass those security parameters. And for us as MSPs, we need to get ahead of the game. So we can start implementing and leveraging a zero trust model and roll out alignments of security best practices to our customers so that we know that we're ahead of the game with the businesses that we're supporting.

The last point, last two points is inforcer's operationalized AI governance and the last bit is build a new MSP service line. Look at what you've got now. Security as a service and maybe we need to start considering governance as a service. Is that a model we should be adopting if we're using AI? Quite frankly, I think it absolutely has to be a topic that we should be a new service that customers should be paying for.

So, let's wrap up. We've got our last episode coming up very soon, the unified MSP, and I want to bring everything back into one single pane of glass for us to manage. But control AI before it controls you. Let's get ahead of the game with the customers that we're supporting. AI isn't coming. Is already embedded in your tenant processing your data and amplifying every security decision that we've made. The only question that really matters is do we start to control it?

Are we controlling it or does it start to control us? These are the things that we need to discuss. We need to have present with our customers and it's all part of that defend, govern philosophy that we're building today. Thank you very much for your time. I wish you all the best and I look forward to the last episode.