Name: Enabling Safe Usage of Microsoft Copilot with Forcepoint CASB
Uploaded: 2025-12-04T17:16:07.350Z
Duration: 58 min
Description: Enabling Safe Usage of Microsoft Copilot with Forcepoint CASB

Transcript for "Enabling Safe Usage of Microsoft Copilot with Forcepoint CASB": What I'd like to do is, quickly recap what we are delivering for Copilot, with CASB solution today. But beyond that, in the next couple of slides, I would also like to cover what, how we at Forcepoint think about AI security. What else is in store? What is the roadmap looking like? So that's what I would like to kind of give you that idea of what our vision is, around AI security. So first, I'd like to start with, what we are delivering with Copilot integration. It's CASB API based integration with Copilot. So that means we are doing agentless scanning of all the messages, your users that are exchanging prompts, getting responses from Copilot. That's what, we have implemented. With that, what customers will gain is, full visibility around the interactions that are happening on the Copilot. It could be direct Copilot chat interface or it could be, Copilot embedded in various apps like Teams, or Excel, Word document, etc. We are able to look at all the interactions through AI based inspection and give you clear visibility around, you know, who's accessing how. And as you all know, CASB is integrated with DLP, so you can extend your trusted DLP policies to identify the types of messages that are being exchanged with Copilot. That's one thing. With DLP integration, what you can, of course, leverage is any sensitive data that's exchanged with Copilot. It gets tagged. You are able to do incident management and forensic analysis. Then the third value that we are getting, of course, is analysis. Then the third value that we are getting, of course, is simplified operations because this is API based scanning. While it's not real time blocking real time actions, you get this visibility around the actions. That is where you can actually plan your enabling of AI across your organization. First, it starts with visibility and then you can plan for more blocking actions that can be covered across various different parts of our Forcepoint portfolio. So this is what we are delivering through CASB API. Now I wanna take a step back and, first of all, give you a bigger picture of how our customers, the customers that we frequently talk to and how the market is looking at AI adoption in general. Quite frequently, customers that request AI security from us, they are concerned about, okay, what is the visibility and compliance aspect? It's not just about prompts and responses, but also an inventory of all the AI applications that are in use within their application. It could be AI agents, it could be MCP servers, MCP clients. How do we get visibility around all of that? Then of course, data risk is at the top of the mind. What is the data that's going into and out of these tools and is that acceptable allowed? Then of course, controlling access. First comes the visibility and then you can say, okay, is this an allowed action or denied action? That's real time control. Then also comes with a shadow AI control. If there are unsanctioned AI apps that are being used, how do you get visibility and how do you block? But underneath the foundation of all of these use cases is actually understanding your data. Even before you think about AI adoption, where customers most frequently struggle is, understanding their data landscape. You know, where is their data and how is the data being accessed by various tools and is the data sensitive? Discovering, classifying, and prioritizing is the most foundational principle. Even before you start thinking about enabling AI, this is where you need to think about where your data is and how is that data classified. That's where Forcepoint's full portfolio like Forcepoint DSPM, DLP, CASB can help you give a broader range of protection, not just API based scanning, but also real time controls once you have identified where the data is. Finally, so, yes, we you are going to see a Copilot and there's a Shadow AI demo next, but I would like to paint a little bit bigger picture, broader strategy on where we are headed. Just a word of caution, meaning legal terms that I have to use whenever I'm showing future solutions is any future solutions marks that I share are subject to change. That's a word of caution. As we get closer to release, we will, of course, keep you posted and have more elaborate sessions in future around how things are shaping up. In this particular vision or visual, what I'm showing is we are thinking about AI security as not just Copilot or not just ChatGPT. The AI landscape is rapidly evolving. Most customers want, us to, you know, detect wherever there are AI tools and, it could be custom homegrown AI tools, public LLMs, or it could be, MCP server. Our MCP servers allowed to interact with certain tools, etc. Our vision starts with creating a specific category called AI security, which will allow you to inventory all the applications that are in use along with the risk levels associated with those applications. Showing what our data is going through these applications, how is the data classified? This is where our DSPM portfolio comes into picture. Then allowing you to take actions on either shadow AI applications that are publicly available, allow deny shadow AI applications through our web security, or detecting agents. Are certain actions taken by agents versus humans? How do we detect agents versus human? Is the agent allowed to take that action on a managed device. So these are some of the areas that we are looking into as we bring together a centralized reporting, centralized policy action, for our customers to easily, you know, understand their AI landscape and take necessary actions. So this is where we are headed. Again, this is conceptual, and we will bring more details as we get closer to the release. With that, I'd like to take, I'd like to give control back to Jen for, demo. Awesome. Thank you, Srita and Niraj, so much for walking through those slides. Some really awesome things coming up for Forcepoint CASB and Copilot. I am gonna go ahead and switch over to a demo. I'm gonna go ahead and share this, everyone. It is about a six minute demo, while it walks through, securing Microsoft Copilot with Forcepoint, a day in the life. Hello everyone. I'm Aditya from Forcepoint. In this video, we will explore how Forcepoint protects organizations from Genai related data risks focusing on Microsoft Copilot while briefly addressing shadow AI. Before diving into the solution, let's walk through a day in the life of typical IT user and see how data exposure can happen silently. A user begins the day browsing multiple GenAI websites like ChatGPT, Gemini, and unmanaged shadow IT tools like Box. These unsanctioned platforms pose major risks because sensitive corporate data can be uploaded without any oversight. For example, the user pastes a confidential engineering roadmap and financial forecast into an open source LLM to generate a quick summary of it. At this moment, the data has already left the organization and no one knows. Next, the user opens Microsoft Copilot, a sanctioned and approved AI tool. He pastes sensitive details including customer PII into a prompt. Copilot summarizes the content flawlessly, but here's the risk. Copilot has no inherent awareness of data sensitivity unless proper controls are in place. Another risk associated with Copilot is uncontrolled exposure of confidential files. Copilot respects existing permissions but its ability to reference enterprise files means overshared or misconfigured Microsoft OneDrive and SharePoint permissions can surface highly sensitive documents to unintended users. So when the user asks Copilot to pull information about a customer from a OneDrive file containing customer data that was accidentally shared too broadly, Copilot instantly retrieves and displays the confidential information, Information that was meant to be strictly internal is now exposed through a simple AI query. Finally, the user downloads Copilot's summarized output and uploads it to a personal cloud service like Box. In just minutes, highly sensitive data has been exposed, summarized, reorganized, and ultimately exfiltrated all without triggering any native safeguards. This entire workflow may look harmless, but every step introduces risk. Shadow AI, sensitive prompts, over shared files, and unmonitored exports. And this is exactly where Forcepoint steps in. As the user attempts to access a shadow AI website, Forcepoint secure web gateway immediately steps in, blocking unsanctioned GenAI tools and presenting a coaching page that guides the user towards approved options like Microsoft Copilot. Now the user tries to open Copilot from an unmanaged device, perhaps with the intention of performing actions without being noticed. Forcepoint CASB in reverse proxy mode understands the device context and denies Copilot access to unmanaged devices. Further, when the user switches to a managed device and opens Microsoft Copilot, they begin typing a prompt and paste sensitive information such as AWS access keys. This is where Forcepoint steps in. With Forcepoint's CASB API support for Copilot, sensitive data like AWS access keys are scanned and detected instantly. The security team can review the details in the DLP incident log, including the sensitive data detected, user and device information, time stamps, the create copy action, and the enforcement taken. This gives complete visibility and control without disrupting the user's work. In parallel, Forcepoint CASB API captures the violation and retains the last fifteen minutes of user activity for audit and investigation, ensuring complete accountability and forensic coverage. Next, the user asked Copilot to pull data from a OneDrive that was accidentally shared externally. Forcepoint CASB API identifies the sensitive content and immediately revokes the risky sharing permission using the configured CASB API policy. This prevents both the user and Copilot from accessing the exposed information ensuring the content remains fully protected. Finally, when the user downloads a Copilot generated report and tries to log in to box to upload it, Forcepoint intercepts the traffic and blocks the upload, preventing data exfiltration at the last mile. Without these controls, sensitive data can leave the organizations in minutes unnoticed. With Forcepoint, every step is monitored, controlled, and secured, empowering safe AI adoption. Thank you for watching this video. Awesome. Alright. I hope everybody enjoyed the demo, and we are gonna go ahead and start with our q and a. So we did have a few early submission questions, so much appreciated to those who submitted those. As always, it is an option when you do register for the event. So we can go ahead and just start from the top, if you wanna go ahead and dive in. So our first question here is how to protect a work environment from illegal actions by using Copilot agent, and how would Forcepoint solve that? Alright. I'll take that question, Jen. So that's a great question. Here, I would like to specifically talk about the Copilot CASB API integration, right, with Copilot. And here, the illegal actions could be interpreted as some sharing sensitive data from, sensitive data with Copilot that customers are not allowed to share. So, what we do, this is exactly what we are targeting through, Copilot release. We scan the data, through Copilot APIs, and we kind of tag it. We send it to DLP for identifying the patterns that you have configured. It could be, sharing fingerprinted data that are not allowed to be shared or it could be purview labeled confidential sensitive data that's not allowed to be shared with Copilot, etcetera. So we give the visibility through the DLP if it matches your configured DLP policies as illegal actions. That is what we would like to flag. So one of the things that I wanna talk about here also is once a message is flagged, because this is API out of band remediation, it's not blocking action that we will be taking on Copilot. Right? It the actions that are allowed for API based scanning are safe copy. And in future in near future, we are also bringing delete action and deleting a sensitive or illegal, action in this case. Currently supported action safe copy, what it does is it actually copies the message over to a quarantine area where administrator can review the message, last fifteen minutes of that conversation to get the additional context, and then can use it for coaching the user or removing access to that user for Copilot, etcetera. So that is how, we propose in terms of, AI, you know, blocking illegal actions. Awesome. Thank you. So the next one here is if a company has their own Copilot, is there a way to filter out if someone's looking for data that they're not actually allowed to access but are trying to get it through Copilot, trying to get a workaround? Can force one alert on that? Is that something that they can adjust for? So when they say if they have their own co pilot, is there a way to filter out yeah. So again, it starts with, right, identifying what data is going into co Copilot. So that starts with our Forcepoint DSPM solution. First of all, mapping out the data and classifying the data as sensitive, labeling the data as sensitive. Then comes, you know, that is where our DLP can use some of the labels, that are there and then, prevent, that kind of data from going into, into Copilot. So so with CASB API, first of all, we scan and show that visibility to the users and then, and then say that, okay. This is sensitive data that's going in. Right? So that's what we, actually are proposing here with CASB API. Awesome. Flagging the user, flagging the messages. Awesome. For the next question is how to block fingerprinted data from leading through Copilot integrated browser sessions. When we say leaving, again, that sounds like a blocking action. What I want, what I'd like to say is CASB API will flag, the fingerprinted data through your DLP policy or fingerprinted data, and there is a policy to flag if whenever there is a fingerprinted data being shared with Copilot, that is what you will get the message. You may not be able to block at that minute, with Copilot at the moment, but it will it all starts with flagging on why someone's actually, sharing fingerprinted data. Another solution that I would like to talk about is if, let's say, that file that's that that contains fingerprinted data has a purview label, we can actually restrict Copilot from accessing certain purview labeled files. Right? So it's it's a solution that customer can implement. And so then Copilot should not be able to access those type of files as well. That's another way to block as well. Alright. So I believe the fourth question is asking about the different versions available of Copilot, so desktop, web, enterprise, and public, And if ForceMind CASB will work with all of them or is there just one in particular? Between desktop web versions of Enterprise Copilot, since we are doing CASB API scanning for enterprise Copilot, we, we should be able to scan messages that are exchanged with Copilot, and that should not be a problem. So CASB API can fully support in an agentless way. It can scan and identify where it is where the sensitive data patterns are observed, etcetera. For the public version, I'm assuming that's kind of like shadowy, unapproved Copilot tools. That is where, our CSG solution or endpoint DLP will come into picture. But then, we are doing more around being able to detect inline real time blocking of data through Copilot. That will be a road map item for us. So that that's kind of coming soon for real time controls. For out of band controls through CASB API, that's where you get great visibility. Desktop or web version are supported. Okay. The next question coming in is around CASB monitoring Copilot through current AI API integrations for Office three sixty five. If it doesn't monitor through the API integration, do they need to perform a separate one, or will the current API integration cover the content for Copilot? So if you're already a CASB API customer, for example, using OneDrive or SharePoint, we do need in order to be able to scan CASB API, for CASB API to scan AI messages through Copilot, we need additional permissions from you. Right? So and additional permissions to scan AI messages. So what we are, what we need what we are doing is Copilot is treated as a separate app here, and we do require customers who are already on Microsoft OneDrive SharePoint scanning, via Gatsby API. They need to reauthenticate to give us additional permissions to scan Copilot messages. In addition, I'd like to talk about a nuance. We consider, OneDrive, SharePoint part of Microsoft Office three sixty five suite of apps. Copilot is going to be treated as a separate app. It encompasses all these apps, but it is treated as a separate app. And that's, partly because we are going towards AI security as a separate category, where we will be supporting lot more apps beyond Copilot. So it will be under a AI security category. So it will be treated as a separate app. Gotcha. Gotcha. And then the last question from our early submission questions is about monitoring sensitive data. So it seems like that's been an issue for this particular submitter. So trying to understand better how they can leverage Forcepoint to monitor data movement in Microsoft Copilot. So I think you talked about that in some of your slides as well and then within the demo, but if there's anything else you wanted to elaborate on. Yeah. This is exactly what this demo has been about, what we are releasing. We are this is exactly the tool in an agentless way to monitor Copilot, via CASB APIs and give you that control on planning how to enable that AI. What type of data is going through? Is the data should it be allowed to be accessed via Copilot? This is what exactly the CASB API Copilot support addresses. We are currently in the early access program. Hope you get to try it. If you need access to this, please do reach out to the PM team and we'll we'll be happy to assist you here. Awesome. Alright. That does enter release submission questions. Again, thanks for everyone who submitted those. And now we go into our q and a. I can see Jake and Shirod have been working right through most of those, but I wanna tackle a couple here at the top. So one of the questions came in is, is CASDI solely for Copilot, and what about other AI tools? As I have covered in my AI security vision, right now, we have CASB API support for Copilot. Absolutely, we are focused on a lot of different tools that go far beyond public or enterprise, tools like Copilot, Chad GPT, etcetera. We are looking into, agents, agentic security as well as MCP type of interactions. So that is, on the road map. It will be coming in the, it is a plan for f y twenty six directionally, and we will keep you updated on some of these exciting, new things that we will be introducing in our portfolio. And then the next question came in. From the user's perspective, what type of performance impact would CASB have? We have several customers who are actually in the early access mode already, and we do extensive QA, performance tests on, you know, CASB, when we are integrating with new applications. In this case, CASB API scanning for Copilot AI messages. So far, our performance has been, you know, as an EPM would say, it's been within the expected parameters. So if there are no additional constraints as such. And if there are any issues, of course, we will be working with the customer individually to fine tune any performance constraints that we see. Okay. Alright. So just checking through the chat here for just a moment, see if we're missing any other questions that haven't been answered yet. I think we've got them all so far. I do see a question coming in around having forced my DLP with Copilot Enterprise. But I don't see the following half of that question. But definitely keep them coming. I think this one. might be asking if do they have access to the Copilot features from CASB? And, Sarita, the licensing is separate for Forcepoint CASB. For Forcepoint, CASDI. That's correct. If, if customer has Forcepoint DLP already, as long as they have CASDI API license, they should be able to use Copilot, and, we can help there if customer does not have the license. Okay. Great. That's something we can follow-up, offline. So I do see, who raised the questions. We can definitely follow-up afterwards, with additional information. Oh, I think I see the correction on the question. I have Forcepoint agents. Right? I I have Forcepoint agents with enterprise Copilot. Don't I already have this functionality? Currently, you can you are probably able to do some features, like without CASB. You're able to probably use Copilot, but really monitoring of all the interactions will come from CASB API based scanning. So, I don't think I mean, if you're able to do that with, if it satisfies with just the agents, there are a couple of issues like Copilot will use WebSocket. And today, not all the interactions are logged and everything. With CASB API, we ensure that we are directly connecting via API and give you broader visibility and identify also whether Copilot is part of Excel app, Word app, or what is the referral app that's using Copilot. So this type of visibility you will be getting from CASB API. Awesome, Thank you for the extra detail there. Alright. So I know we got through the questions so far, but definitely keep them coming. Alright. We'll give it a couple of minutes. Typically, we don't get any new questions coming in. We don't mind ending the session early, but I wanna make sure we have plenty of time for those who are still putting their questions together. There's an interesting question here. I know it has already been answered around using Forcepoint SWIG. But I wanna see, sir, if you have any other comments here. So the question was, I wonder if there is still a way from a DLP perspective to prevent data exfiltration if we set up a proxy exception for the personal storage backup. The response was that you can use Forcepoint Suite to accomplish this. I wasn't sure if there's anything else we could add to that. Okay. Sorry. I'm reading that question. I wonder if there is still a way from DLP perspective to to prevent data exfiltration for personal storage backup. You can use Forcepoint, Sui, to accomplish this. So I'm just trying to understand the question. We set up a proxy exception for personal. So I think what the customer is saying here is, they they have a proxy exception for personal storage and backup. And, so so so that means that through SWIG, perhaps it's not going through proxy. The personal storage and backup is not going through proxy, cloud proxy, and how could they prevent any data exfiltration attempts. I think we should be able to do it via endpoint DLP. Let's say that data is actually going into, some of the sites. But the problem with endpoint DLP is also that, you know, you can't say, hey. This is just personal, personal, OneDrive versus, enterprise OneDrive. So it it will treat everything the same. So I think I think once you have a proxy exception, it really cannot be controlled. You have to like, whatever approach that you're using to security, right, you may you have to make sure that your exceptions are configured accordingly, so so that we are able to inspect one way or the other. So if you have an exception, you have a risk, and we need to understand why that exception is there and, why why we are getting that blind spot. That would be my really very high level answer. If there is a problem here, I would really like to, you know, double click and, perhaps get through a support ticket. Perfect. So a question came in, about the again, just about the licensing model for CASB API scanning. Mhmm. Oh, CASB API scanning, if you are already a DLP customer, we have a good news. We integrate CASB API through DLP, and there is a SKU, available for CASB API, enabling CASB API on your existing DLP. It can be Cloud DLP or on prem DLP. And the the SKU covers, entitlement for up to three applications. And, one thing that Nuance that I was mentioning earlier, you can have Microsoft scanning, OneDrive, SharePoint as one application. Copilot will be treated as a separate second application. So as long as you are under that three app limit, that's what we provide you. And if you need to buy if you need to protect more apps like Slack or other applications in future, you can always buy more from us. Perfect. Alright. We did get the fifteen minute reminder. We are coming up to the top of the hour. So if there are any additional questions coming in oh, here we go. During the let me open this up. During the configuration of Copilot and Office three sixty five, this user is already they already limit the data to be shared within the organization level. And while using Copilot, how can the data how our data will get globally declared and risk is measured? So access will be provided based on usage, insight into how policy is first pick. One level we can achieve it from Forcepoint DLP where the destination is marked with CASB. So this is pretty, detailed question. I'll try to do my best, but, if we need to have additional discussion, I'm happy to have it. So just reading the question, during the configuration of Copilot and Office three sixty five, we already limit the data to be shared within org. While using Copilot, how our data will get globally declared and risk risk is measured. So to me, right, security is not a one time process and especially with organizations. There will be a lot of users. There can be, you know, a lot of blind spots that they that get created. So if I'm understanding the question correctly, this is about exposure. The actions that you took already are, probably, you know, keeping you safe in terms of your Copilot enablement goals, but it is also about is there any configuration drift that's causing new exposure. So and sorry if I'm not understanding this question right, and I'm happy to spend few more minutes. Alright. So this is about getting that continuous visibility. Is there any additional data that's getting shared that will surprise you when you see it through CASB? So that's what this is about and what what controls that you can, take. Right? If you if CASB detects additional data, that's where you need to go back and implement additional restrictions. So that's how we see, CASB coming into picture for that continuous feedback, continuous monitoring, and making sure that you are able to achieve your goals. I'm just kind of giving, a quick read on the last part of the question. The level of access will be provided based on usage. Kindly give us insight for how policy is first pick. So I think what, again, this audience member is saying is, yeah. We do. The destination will be CASB API from Forcepoint DLP. That is how we scan Forcepoint DLP. CASB first scans once you configure Copilot within CASB. Right? CASB starts scanning and the Forcepoint DLP is our backbone verdict engine and we send it to, CASB any data that we find from Copilot. That's how the policy selection happens. And if CASB or DLP says, yes, this is fingerprinted data or, yes, this is data contains PII that should not have been shared, that's when we flag that and you can see that in the unified incident management console on DLP side. I hope that answers, but I'm happy to like, if you're interested in enabling Copilot for your, security, maturity purposes, you know, I'm happy to have a discussion separately on the road map. Appreciate it, Sarita. So we've got about ten minutes left. I am gonna go ahead and push out our survey. Again, we do read every response. We look at all of these. So any feedback you have anything you'd like to see going into FY 2026. We are in the planning stages right now for what our schedule is gonna look like next year. So love to take any feedback you might have, across Fort Smith Cassidy with other products you might have in your portfolio, happy to hear it. Alright. Go ahead to q and a. Alright. So next question just came in. How are sites with AI recognized? Is there a global list that is updated from time to time, or do administrators have to create their own website lists? Absolutely. This is a great question and detect specifically addresses the web security part of the solution. Right? We do have a huge shadow IT database and we have a category for AI apps that get frequently updated, and that's how we recognize. And in addition, custom customer can also create custom categories and keep adding, but our, we are kind of investing in this area to keep, the list updated. And, this is a massive area as there are tons of applications coming in every day. We keep this list updated through our web security, shadow shadow IT portfolio. Alright. We've got a few more minutes here. So any last remaining questions that folks might have? Great questions coming in on today's office hours. Alright. Can you clarify blocking functionality as you mentioned tagging instead Yeah. So in, with CASB API Copilot support that we've? introduced, I'd like to once again clarify that API is out of band remediation. It will allow, you to look at the messages, classify the messages through DLP policies, and safely copy the message that is offended some policy that matched the policy. So for Copilot specifically, the blocking action is not available yet, and that needs to be in the road map. That's what I was talking about. For other apps like chat GPT, we do have blocking via endpoint DLP, CSG, etcetera. So that blocking functionality is available. But for Copilot, we don't support blocking yet, and that is on the road map. I hope that answers. Awesome. Alright. Any last minute questions that wanna come in before our session is over? Alright. We'll give it another minute or so. And don't mind ending a few minutes early, so we're right on time here. Again, appreciate everyone joining. The questions you've asked both during our chat today and the early submission questions, really great stuff. Sreeda and Neeraj, appreciate you both joining today, sharing our future roadmap for CASB. Looking forward to what you come back and share in f y twenty six. And then, of course, Jake and Sharga behind the scenes answering all of your questions. I think a follow on to the blocking question is, do you have a timeline of when you do hope to release that functionality? Some point in f y twenty six, but if, again, we can, be more specific and these timelines, are subject to change. So I'd like to have a more detailed road map session with the customer. So happy to do that. So we are actually it's on our road map, for f y twenty six, and we'll be more specific as we as the work gets planned. And it's on you know, it has a commitment process and everything, that it goes through. Perfect. Perfect. Sorry. I guess it was just the Copilot API features in general. Oh, okay. Sorry. Great. Copilot API. Yeah. Copilot API, we have already released it. It's in early access mode. It's available right now for you to try get that great visibility. And, again, our early access team, they are super excited to be helping customers. They give that, hand in glove care for the customers to make sure that we capture your feedback. We are able to set it up correctly and everything. Please reach out if you're interested. We are able to set it up fairly quickly for you. Awesome. Alright. Just about just just under five minutes left. Alright. Great questions coming in today. I'll give it another minute or so here to see if anyone has any last remaining questions before the session is over. Alright. I think we've got the last remaining questions here. Again, thank you all so much for joining. Really, really appreciate your time and attending our Forcepoint office hours with us. Whether this is your first or your tenth office hours with Forcepoint, we do appreciate you here. Like I said, looking forward to FY 2026, what we haven't planned, not just for Forcepoint magazine, but for all products for office hour series. So stay tuned for more coming. Awesome. Thank you all so much.