Microsoft Ignite 2025: How Microsoft Purview Drives Data Security in the AI Era

Microsoft Agent 365 Will Help Secure Data Being Used by AI Agents

At the event, Microsoft introduced several new security features to aid in protecting data being used for AI. In one session, Charlie Bell, Microsoft’s executive vice president of security, said, “We’re seeing a 4x increase in click-throughs when phishing is AI-automated. Leaders are understandably worried about data leakage, and organizations are using more tools than they keep track of.”

“AI also creates new attack surfaces that are layered on top of traditional threat vectors, manipulation from web and data sources, poisoned or unvalidated AI models, and agents coerced into becoming double agents,” Bell said. “We have to protect these surfaces comprehensively.”

One of the new offerings Microsoft presented to address these concerns is Microsoft Agent 365. According to a recent company blog post, “By 2028, businesses are projected to have 1.3 billion AI agents automating workflows. Most organizations don’t yet have a way to observe, secure or govern them — if not governed, AI agents are the new shadow IT. Microsoft Agent 365 enables you to observe, manage and secure your AI agents, whether the agents are created with Microsoft platforms, open-source frameworks or third-party platforms.”

WATCH: See why agentic AI introduces new security concerns around identity and access management.

Joy Chik, president of identity and network access at Microsoft, told attendees that Agent 365 “takes the same infrastructure that you trust to manage and secure your people and extends that to agents and their unique needs. So, this makes it easy for security and IT to collaborate on protecting and managing agents.”

Agent 365, Chik explained, starts with a registry for a complete view of every agent in an organization, including agents with identities in Entra Agent ID and the agents registered by the organization, as well as shadow agents.

Once all agents have been identified, the next step to securing them is access control. “Agent 365 enforces the least privileges and applies conditional access policies that understand context and the risk,” Chik explained. “Visualization lets you explore the relationships between agents, people and data so you can monitor their behavior and performance in real time to see their impact.”

Microsoft Announces Enhancements To Provide Greater Data Security

Talhah Mir, partner group product manager for Microsoft Purview Data Security, spoke at a related session on data security as a holistic discipline, noting that there are three dimensions to consider.

“First, how do you look at and manage the risk to your data at rest? Then, how do you manage your data in transit, or the risk to your data in transit? And last, certainly not least — because data doesn't move itself, people move data — you’ve got to look at the user risk as well,” Mir explained.

Microsoft has layered data security into several offerings to address each of the dimensions Mir mentioned: Microsoft Purview Information Protection for data at rest; data loss prevention for data in transit; and finally, insider risk management for user risk. “And they’ll come together and light up these amazing capabilities around adaptive protection and even have new capabilities, like data security investigations.”

Mir was excited to introduce the company’s Data Security Investigations, which he called, “one of our newest babies out there right now. This is really the first AI-powered, born-from-the-ground-up solution to help you investigate your AI or your data risks using the power of LLMs.”

Click the banner below to learn how enterprises are unlocking artificial intelligence’s potential.

Developing a Holistic Foundation With Data Security Posture Management

“The idea is, you want to manage the risk across those three dimensions,” Mir said as he began to explain Data Security Posture Management. “You don’t necessarily want to have to worry about stitching them up across those solutions. You want that objective, you want the outcomes, you want to achieve that in one place. That is DSPM.”

Data security can be overwhelming, given that every organization’s data profile is constantly emerging and evolving. There’s always new data being created, and it’s difficult to sift through so you can identify the key data items and know what you need to protect and how you need to protect it, which will allow you to accelerate your AI adoption, Mir said.

Rob Sobers, chief marketing officer at Varonis, shared his experience and thoughts on data security when using AI. “With AI, it is a data security problem, and AI is built by design to consume all of the data in its pathway. If you have Copilot deployed, it’s using the user’s pass-through permissions. If you have an agent, the agent is more valuable, as it can consume more data, but you have to have the right controls to make sure it’s only the right data.”

Mir said, “Oversharing is not an AI problem. It’s a data security hygiene problem that we need to go fix.” And sometimes, that fix can’t be made easily. As several of the speakers said, data security can be “messy.” But it can be easier to handle with a trusted partner or through a managed service provider.

Sobers announced that Varonis is partnering to deliver better visibility, especially for regulated data. “What we’re doing with Varonis for Copilot and Varonis for ChatGPT and other AI systems is making sure that only the right data gets in the right hands.”

To learn more about Microsoft Ignite, visit our conference page. You can also follow us on the social platform X at @BizTechMagazine.