AI agents & Security for AI

Hi everyone and welcome to the 12th edition of Heart of Security! It’s been so rewarding to share this with you for the past year and I look forward to sharing even more from the security world in the months to come. As I’ve talked about in the past, I am an AI optimist. AI is an incredibly consequential technology with the power to elevate human potential. And with every stop we make as part of the global Microsoft AI Tour, I’ve had the amazing opportunity to see this in action – customers and partners innovating with AI and embracing change in inspiring ways.  

And speaking of inspiring, it feels remarkable to me that just one year ago, we introduced Security Copilot to the world and this week we announced the addition of 11 agents in Copilot that will strengthen protections and increase productivity. You can learn more about the agents we are bringing to Security Copilot in my blog and check out this great video that explains the immense value they offer.  

Our innovation in AI doesn’t just apply to using AI in security, it also extends to the way we help organizations secure AI deployments and use AI safely. According to a recent Microsoft whitepaper, 95% of organizations surveyed are in some way planning to or already using and developing AI. Two thirds (66%) are going even further, developing AI of their own. And the overwhelming majority report doing so for a single reason: innovation. But how can organizations ready themselves for AI? Here are four steps I recommend:  

Before you introduce AI to your business, it’s so important to put security first. Microsoft has the privilege of seeing tens of thousands of customers using our AI apps and building their AI on our platforms, which provides us with a tremendous amount of learning on how to secure and govern the AI our customers use and build. It’s one of the reasons why, as a leader in both the AI and security spaces, Microsoft has placed the highest priority on doing two things as we introduce AI tools.  

The first is to secure our AI tools, products and platforms, so we can ensure they adhere to the strictest guidance regulations and follow best practices and risk management frameworks when it comes to privacy or data access. To that end, we have developed our Responsible AI Framework, our Secure Future Initiative, and our Privacy Principles, all parts of the commitment that we’ve made for trustworthy AI to ensure that users are protected at every layer.  

The second is to provide organizations with comprehensive security tools so they can better protect their own heterogeneous use of AI. We know and expect organizations won’t be using our tools alone. On average, companies are using or developing an average of 14 GenAI apps. There are literally hundreds and hundreds of apps being developed in GenAI, and any approach to comprehensively securing AI must account for that. Consequently, our approach towards security for AI has been anchored on four steps:  

• Prepare your data. I cannot emphasize how important preparation is. Before implementing AI, you need to make sure that your data is classified with all the right sensitivity labels and governance policies applied. You must also have the right policies for your organization: ways to consistently, reliably and automatically classify information within your data estate to help prevent possible overshared or leaked information.  

• Discover your current AI use. Many times, people don't even know what AI apps are being leveraged at work and whether they are authorized for use. In all, 78% of AI users bring their own AI tools to work, sometimes without the knowledge of the IT or security group within an organization. You need a way of knowing what “shadow AI” applications are in use at your organization. This visibility into AI app usage should also include visibility into potential risks like users accessing data beyond their authority, leaking sensitive data, or even creating noncompliant content. 

• Protect your users and their data. Adopting insider risk management (IRM) tools can help detect and respond to the inclusion of sensitive information in AI prompts or responses and alert security teams of potentially risky AI use. For the best protection, combine those alerts with adaptive data protections that can automatically adjust  privilege access management to prevent and mitigate potential data breaches.     

• Govern your data estate. To protect sensitive data, teams should configure data loss prevention policies to help prevent users from accessing or sharing sensitive data through AI apps. This is a significant challenge for any business; EU AI violations can cost companies as much as 35 million euros, and 62% of business leaders say they don’t understand the AI regulations that apply to their sector. Make good data governance and compliance easier by adding new regulations to your compliance management so you know what regulations you must adhere to.  

Any organization interested in leveraging GenAI, whether it’s a ready-to-go assistant like Microsoft Copilot or one developed in-house, should take these steps before and during adoption. With proper preparation, discovery, protection and governance, you can ensure your organization will be able to minimize and mitigate AI risk and get the most out of what AI has to offer. 

I had the absolute pleasure of sitting down with my colleague and friend Dorothy Li to talk about how AI agents are changing the game for security. Dorothy started at Microsoft in 2023 with the goal of evolving AI security offerings for the betterment of the industry and the people they keep safe and secure. She has been working in the technology industry for nearly three decades and says that “Security is the defining challenge of our time and is one domain where Gen AI can have a transformational impact. Knowing what we can do with the latest technology to keep our customers (and their customers) safe makes me appreciate how critical our team’s work is and the impact we can have by making security teams more effective and efficient. There is nothing like knowing your work is fighting bad guys to stay inspired!” 

On Monday, Dorothy helped me announce the exciting news that Microsoft expanded Security Copilot with six security agents from Microsoft and five security agents from partners, which will be available for preview April 27. AI agents mark a major transformation for defenders and in SOC operations and operate on a speed and scale that cannot be replicated otherwise. 

The intersection of AI and cybersecurity is constantly evolving. Innovation and collaboration with our ecosystem of customers and partners is a necessary and rewarding component of enabling AI that has the power to elevate human potential – and that’s part of what makes it so exciting. Dorothy says, “Every day is different. We will never get to a point where we have solved every problem and get to declare our mission complete. Every day presents new developments and new challenges which require that we stay current on the latest research, understand the rapid changes in the threat landscape, and how we can close the security gap that often exists.” Dorothy shares that the most important skill for someone to succeed in the AI or cybersecurity fields is “insatiable curiosity” and that those looking to get their start in cybersecurity don’t have to look far to make a meaningful impact. Whether you’re in product management, translating research into useable features to solve real customer problems, or marketing creating messages that resonate with customers, Dorothy says it takes a village to succeed in security. 

Generative AI is here to stay and it’s important to embrace AI in your daily life and current role – both to stay ahead of the curve and to make your life easier. Dorothy says, “There is fascinating research coming out and complex applications but there are also myriad real-world applications for AI. Challenge yourself to find ways to use AI in anything you need to do today from automating a repetitive task at work to figuring out what you should make for dinner.”   

When I asked about her favorite prompts to use at work, Dorothy said, “‘Draft…’ since AI is great for writer’s block! I can quickly go from a few ideas in my head to a first draft. The value of AI is really in freeing up the human mind to do higher level work. In this case, once I have a first draft, I can spend my time on the more important work of thinking through action steps to move work forward versus just getting ideas out of my head.” 

I couldn’t be more excited for all the ways AI is making life easier, from everyday tasks to how agentic capabilities in cybersecurity are keeping us safer online. 

• I am so thrilled about the reception to our announcement of agents in Security Copilot and new innovations for securing AI! Read more about our Secure news in The Verge, Axios and CRN. I also discussed the news in an interview with Sky news! 

• If you would like to learn more about our AI agents news, I invite you to our Secure digital event on April 9, an in-depth digital security event with a focus on security for the age of AI. 

• I’m so excited about the launch of Microsoft Purview’s Data Security Investigations (DSI), a new generative AI-powered solution that helps data security teams quickly understand and mitigate risks associated with sensitive data exposure. Learn more from Herain Oberoi. 

• It’s been both enlightening and heartwarming to hear how organizations and people are using AI to expand human potential. Learn how universities are tapping students and AI to fight the growing threat of cybercrime from Microsoft Source. 

Something that recently inspired me is a book I am reading called The Power of Moments which talks about how we can create intentional moments to inspire change, create joy and drive impact. 

A quote I love: "When you become comfortable with uncertainty, infinite possibilities open up in your life.” – Echart Tolle