Video: A CISO Guide to Securing Copilot and the Future of Work | Duration: 1808s | Summary: A CISO Guide to Securing Copilot and the Future of Work | Chapters: Welcome and Introduction (26.27s), AI: Productivity vs. Risk (166.03s), AI Adoption Risks (326.905s), AI Native Security (630.875s), Implementing AI Security (957.27997s), Recap and Conclusion (1446.565s)
Transcript for "A CISO Guide to Securing Copilot and the Future of Work":
Hello, everyone, and welcome to today's session. I'm Neeraj Knight, the senior product mark marketing manager at Forcepoint, and I have with me Sarita Chaddalawada, the senior director of product management at Forcepoint. Sarita, welcome. Hi, Neeraj. Looking forward for this discussion. So in today's session on c so guide to securing Copilot and the future of work, we'll explore how security leaders such as yourself can securely enable AI applications like Microsoft Copilot without compromising on innovation or productivity. We'll share insights from the field and show how Forcepoint helps organizations become AI ready, balancing speed with security and compliance. Now let's begin with the simple truth. Every organization today is on an AI journey. Generative AI, agentic workflows, AI driven SaaS applications, they are all part of our daily lives. We are all experiencing this incredible shift in AI adoption. But alongside the benefits, we are also seeing the fastest wave of data consumption and risks that we have ever encountered. Now AI consumes the data everywhere, in applications, in data centers, and also through people like us through the prompts that we enter. And here is the kicker. About 89% of the enterprise AI usage is invisible. It's happening outside the control of your IT and security teams. And this means that the sensitive data, it can flow through prompts, it can flow through your plugins or agents without you even noticing it. Now the one thing that AI needs to operate effectively is data. AI learns from data. And because your data is everywhere, your AI deployment, it is as safe as your data. So the real question isn't how we use AI, but what data is it allowed to access and consume. So as a leader, you need to ask yourself, am I absolutely sure that my AI deployment is accessing only the data it should and nothing beyond that. With that thought in mind, I'll hand it over to Sarita to explain this paradox that we're experiencing when it comes to AI adoption. Sarita, over to you. Absolutely, Neeraj. That is the right mindset that every leader should go with. GenAI is the biggest productivity unlock we have seen in the recent years, But some capabilities can but some of these new capabilities can create, hidden risks for our customers. Often, well meaning, well intentioned employees just trying to get their work done can expose data unintentionally to the generative AI tools that they're using. That's the pair paradox that we are talking about here. AI can be the biggest force multiplier to improve your productivity, get work done, but it can also be the fastest data exfiltration tool if you don't add proper guardrails to the usage of AI. So the answer to most of the customers that I talk to is in is it it's not to slow down the productivity, we are experiencing. It's to secure the workflow so that end to end, workflows are secure. We are not stopping the progress of innovation, and we are able to do it safely. We are able to allow the usage of AI safely as we do this. I think that's very well thought. AI can be your data's force multiplier or the fastest data exfiltration tool. And let's ground this with what we are seeing with the adoption of Microsoft Copilot. The momentum around Copilot is real. More than 70% of Fortune 500 companies, they have deployed Copilot in some form or the other, and its usage has seen massive growth year over year. Industries from financial services to health care to telcos and and tech tech services, they are leading the charge when it comes to the Copilot adoption. But at the same time, there are also concerns emerging. According to a survey by Gartner, 57 of the organizations who participated in the survey, they have limited their Copilot rule rollout to only the low risk and the trusted users. They are they still don't trust a complete organization wide rollout of Copilot. And around 40% of organizations, they have delayed their Copilot rollout by three months or more. Now this indicates that despite strong momentum in AI adoption, there is still some resistance in the industry because of lack of clarity on how secure the deployment will be and how AI is going to handle their data. Now, Sarita, you have been interacting with the field almost on a daily basis. Now can you share some of these leading risks that you are seeing our customers face when it comes to the adoption of Copilot? Absolutely, Neeraj. We as I talk to many customers, their concern is more around how do they enable Copilot or chat GPT safely for enterprise usage so that they can actually, they can actually benefit from the productivity that AI tools offer without creating risks for their data. So instead of looking at all the risks as isolated risks, we should look at the cause and effect relationships. Let's look at some of these causes, what makes our IT leaders be concerned about AI usage. Right? The number one is uncontrolled data exposure. For example, lack of visibility into data that goes into, say, Microsoft apps like SharePoint, OneDrive, Teams. If there are overly permit, permitted files and they don't have visibility into it, Copilot can access these over permit permitted files. So that's uncontrolled data exposure is one concern. Then fragmented visibility for administrators. For example, administrators may not have complete visibility into what kind of data is going in, what kind of sensitive prompts the users are entering. So that's the second risk that they have even when using the approved tools like Copilot. Then there's always a risk of insider misuse. What if someone allows Copilot on personal device? And because of overly permissive files, what if, a risky employee is trying to download sensitive data to on their personal devices? So that's the misuse that we are talking about. And then there is always a risk of shadow shadow AI. What this, what this brings is, today, it's very easy to use tools that are not approved because employees may be downloading unapproved, unauthorized tools, the lack of visibility on which AI tool is being used for, and then what is the risk that the tool carries. Right? There's if there's no proper vetting process on the tools, then definitely the IT leaders are concerned about enabling any AI for their organization. So all of these risks, actually kind of lead to two effects. One is compliance gaps. Obviously, the lack of control on the tools and what data is going in are is the data safe to go into that tool. It it actually kind of creates compliance gaps for our administrators or IT leaders. And then always there's a risk of data loss. If the tool is not properly vetted and it is leaking sensitive data, then that actually creates the risk of data loss. So the key takeaway here is absolutely the framing of this problem matters to our administrators. They need to think of this cause and effect relationship so that they are able to manage the risks proactively, the causes proactively, and prevent the ultimate penalties like compliance gaps and losing the organizational data. So this is how the IT leaders, need to need to be thinking whenever they're enabling AI for their organization. Now let's, for example, look at, a typical internal deployment of Copilot and what risks it can bring in. Right? Here, I have an example of an employee, John, who's asking a question to Copilot. Show me the salary details of the team. He could be in the HR team. So show me the salary information. This is obviously sensitive information. And if the salary details file is not stored properly, if it's actually, exposed publicly or externally, then Copilot can easily access and pull the data from HR, OneDrive, or SharePoint. Copilot then dis, displays the information for the administrator, for the user in plain text, and the user can then copy that, onto their personal device or personal cloud. So this is the type of risk that IT leaders, are always concerned about, accidentally exposing data to their AI tools that can then be easily retrieved by, any employee within the organization. Yeah. I think that's a fantastic example. And here, one more aspect to look here is, the way an employee can pull the data and the way the data is is moving through the different channels and the different systems, it can actually change its form. For example, the employee can pull the data from a database, which is a structured data. You can upload it to a file. It becomes a nonstructured data, and it can exfiltrate it. So I think that's also a a key aspect to look at when we, enable access for AI tools such as Copilot to all our sensitive data. Absolutely. And without these guardrails, you know, all your confidential files, they can turn into your exposed outputs. Now to combat the security and governance challenges that are introduced by Copilot and other AI apps, what we recommend today is an AI native security approach. And there are two dimensions to it. Right? So there can be security for AI, protecting sensitive data, enforcing compliance as the AI tools that interact with the enterprise systems. This will this will ensure that any wrong data does not fall in the hands of the AI. Data is already protected. Data is already compliant before it even is, is accessed by your AI systems. And next is AI for security. Now the only way to keep up with AI is with AI. Use AI to continuously discover and and automatically classify your data. You can use AI based recommendations to define your policies to protect your data in real time, all without disrupting your productivity. And it's not one or the other dimension. Both need to work together. Enterprises, they need both to to be Gen AI ready. So so, Sarita, let me bring you back into this discussion. So can you explain how Forcepoint bring this brings this AI native security model to life? For example, if I'm a CSO, how can I be assured that my data will remain protected even if the risks on the data change dynamically? Or, how can how can I get a peaceful sleep at night knowing that I have a remote employee who is using Copilot at work? Right, Neeraj. So with Forcepoint, what we deliver to our customers is a single unified platform where enforcement automatically adapts based on user behavior, and it gives deep visibility as well as, forensic capabilities to the data administrators. Our philosophy is around this five fold approach. We discover and classify the data. That is where the that is the first powerful tool that we offer to our administrators. We use AI mesh based discovery and classification where we can scan structured or unstructured sources to identify data accurately with minimal noise. Once the data is identified, the next thing is prioritizing. Is this sensitive data? Is this not sensitive data? And how much of sensitive data, which will actually, allow administrators to assess the attack surface that they have or exposure surface that they have. So once we discover, classify, and prioritize, the next steps are already around remediation and protection. Here is where, the dynamic controls that we can apply through behavioral and con context of the controls come into picture. We can enforce policies dynamically based on user behavior and activity, context. If, for example, a file content changes, we can rescan and reclassify the files and rescore the risk that is associated with this with this file. So this is where adaptive risk scoring comes into picture. So, overall, this process of continuously discovering, classifying, prioritizing, and continuously applying, dynamic controls to the files. It is, actually a full, life cycle of how we how we control the risks. And this can be applied to either data at data at risk. You may have files that are from ten years ago not touched but exposed. Or it can be a data in use, like any a user is actively modifying a file and how do we get the latest up to date changes on the file. Or it can be a data in motion, like data is going from one place to another. How do we intercept that and then make sure that, you know, the risks are contained? So this is a continuous process that we apply to any type of file, that, we can let us, review the key recommendations here. Here is a blueprint that we typically give our customers who are concerned about enabling AI safely. So, the first step is to get the visibility and compliance assurance. We give full visibility into prompts, responses, file attachments, and capture sensitive conversations and generate compliance ready reports. Then the next step is to mitigate data risks. What what we mean by that is if data is overexposed, then remove the overexposure, revoke access to sensitive files before Copilot can summarize or expose them, and apply real time data loss prevention controls to prompts and outputs that are going into, AI tools to stop the data leakage. The next is controlling access, which is around, creating user and device aware policies to allow or block Copilot based on posture, based on user risk, device posture, and device trust. For example, if you do not want certain users to access Copilot, then how do we allow or prevent that? If you do not want your organizational users to access Copilot from personal devices, how do we prevent that? And the last, use case, in this list is, of course, shadow AI controls. This is about detecting and blocking any unauthorized tools that you do not want to allow your organizational users to use. So these are the four use cases, the key use cases that we, give to our customers, so that they can enable AI safely. What these use cases actually deliver, these are the benefits that these use cases, they give the focus in terms of how or how customers can enable AI. So the risks that they can reduce are, you know, they can prevent exposed sensitive data, through Office three sixty five or any other SaaS application. They can streamline their compliance. For example, they can boost trust and adoption while meeting regulatory expectations. Right? And then they can increase the productivity. They're not always saying no to their employees when it comes to AI usage. They are able to safely roll out AI, with this by focusing on these four use cases. And, ultimately, they are able to cut costs both in terms of productivity gains as well as in terms of any penalties because of compliance gaps or exposure. They're able to, address cost reductions as well. So this is how, most customers are looking to adopt, or enable AI tools across their organizations. Thanks, Anita, for sharing the use cases. Now why don't we show how, Forcepoint secures the data in Copilot through a quick demo of our product? The first point, it can protect your Copilot data through or data in Copilot through both API based and inline controls. But for this particular demo, we'll be focusing on our API controls to discover any sensitive data that are entered through prompts in Copilot and what action can be taken when our, policy detects any such violations. Let's begin by seeing how easily a user can unintentionally or intentionally expose sensitive data through Copilot. For example, I'm entering a script to deploy VMs in AWS that includes an AWS access key. This action could publicly expose the key, violating company AI use policy. Now let's see how Forcepoint DLP helps monitor exposure of sensitive information. With Forcepoint's proven DLP integrated into Copilot, sensitive patterns such as AWS access key or Google API keys are instantly detected. When a match is triggered, the policy securely saves a copy of the conversation into a predefined location in OneDrive. This gives security teams complete visibility into user intent and allows them to provide targeted coaching to the user. Further, with DLP reporting, we can clearly see which policy was invoked, the AWS access key entered by the user, and the action that was taken. Let's switch to Forcepoint insights where we are presented with a powerful GenAI driven dashboard. This dashboard provides a holistic view of user activities and the actions Forcepoint has taken to prevent sensitive data exposure. Further, we have dedicated dataset for Copilot showing granular details like file name, file path, DLP pattern, interaction type, session ID, and refer app. These details helps connect the dots across user activity and applications. Finally, let's navigate to OneDrive to access the safe copy of the Copilot conversation created by Forcepoint DLP policy. This safe copy gives security teams clear visibility into user intent and helps them take appropriate action. Sarita, would you like to, provide your recommendations on the first steps that organizations can take towards AI native security? Alright. Now that you have seen the demo, I would like to quickly recap, on what steps leaders can take towards enabling AI safely. Here is a four step plan for a quick recap. First is leaders should always assess their generative AI risks. That means they should be able to map how their data flows, what kind of tools their organization the organizational users are using, like shadow tools, and assess the exposure point. Is are is their data overexposed? The second is, establishing an AI security council where you have key stakeholders across the organization. For example, your legal, compliance, security, IT, and business teams together, and they are able to form informed decisions around which tools are safe, which tools cannot be allowed, etcetera. Then the next step is to build AI native posture. This is around discovering, classifying, the data that is going in and out of AI tools, fixing any oversharing. Like, if the data is overexposed to begin with, how do we fix overly exposed files? And then, then this leads to the final, plan, which is around continuously monitoring and improving on the plans. Like, getting the visibility around how the data is flowing and removing exposed data. And then doing it on a continuous basis is what gives our leaders confidence in terms of, allowing safe usage of AI across their organizations. Thank you, Neeraj. Thank you, Sarita. And let's quickly recap before we wrap up. Now Copilot is seeing a huge momentum across industries. There are risks associated with Copilot, but those are absolutely manageable with a few key steps. First and foremost, we need to have a secure AI native risk posture by discovering any data that is getting accessed by Copilot, classifying the data, and ensuring that we take proper remediated steps to mitigate any risks around the data. For example, fixing any data that is being overshared with Copilot or, fixing any raw data that may be existing in your OneDrive or SharePoint folders, which Copilot is is accessing today and your IT team is not aware of it. Along with that, you can extend your security controls, beyond Microsoft three sixty five, the native controls, to protect your data outside your Microsoft three six four deployments. And you can enforce continuous adaptive protection to protect your data everywhere. Now if you're ready to make Gen EI safe for your enterprise, we would strongly encourage you to book book a demo session with us, where we'll show you how through our self aware security model, we we can protect your data addressed in use, and in motion. We have also made a practical guide available for you to help you take the first steps towards building a gen AI ready enterprise. And with that, we have come to the end of today's session. Thanks, everyone, for tuning in, and thank you, Sarita, for joining me in this insightful discussion. My pleasure. Thank you. Thank you, everyone.