Jeff Margolies, Chief Product and Strategy Officer at Saviynt – Interview Series

Jeff Margolies, Chief Product and Strategy Officer at Saviynt, leads product management, corporate strategy, technology partnerships, and corporate development, working with customers, partners, and teams across the organization to shape the company’s long-term product vision and growth roadmap. He brings more than 25 years of experience in identity and cybersecurity, having previously served as Saviynt’s Chief Strategy Officer and held senior leadership positions at Mandiant, Deloitte, and Accenture. At Mandiant, he oversaw strategy, strategic partnerships, and corporate development, while his earlier career included helping establish and scale Accenture’s identity and access management practice into a global organization of more than 700 professionals. Margolies holds a bachelor’s degree in Optical Engineering from the University of Rochester and a master’s degree in Electrical Engineering from the University of California, Berkeley.

Saviynt is an AI-powered identity security company focused on helping enterprises govern and secure access across human users, non-human identities, machines, applications, infrastructure, and AI agents. Its Saviynt Identity Platform brings together identity governance, privileged access management, application access governance, external identity management, non-human identity security, and identity security posture management in a unified cloud platform designed for cloud, on-premises, SaaS, IaaS, and hybrid environments. The company has increasingly positioned itself around the needs of the AI era, including tools for securing AI agents, governing agentic workflows, and enabling natural-language identity operations through offerings such as Saviynt MCP Server in AWS Marketplace.

As Chief Product and Strategy Officer, you have a front-row seat to how enterprise identity security requirements are changing. How has the rise of AI agents altered the way organizations need to think about identity, access, and governance, and how did those shifts influence Saviynt’s development of Agent Access Gateway and Intent-Aware Runtime Authorization?

AI agents change identity security because human and machine identities operate in very different ways. It’s no longer just managing people logging into applications. Organizations are managing software that can reason, make decisions, and act autonomously across systems. I think of agents as the next generation of applications. After mainframe, client-server, web, and SaaS applications, we are now entering the era of agentic applications.

The fundamentals are still the same. Organizations need to know who owns the agent, who can use it, what systems it can access, and how its activity is governed. Machine identities change the speed and scale. A business user can now create an agent in minutes, which means governance must move just as quickly. That is what led us to develop Agent Access Gateway and Intent-Aware Runtime Authorization — to help enterprises adopt agents with the right access controls, accountability, and guardrails from the start.

AI agents are rapidly moving from experimentation into production environments. How do autonomous agents change the traditional identity and access management model that enterprises have relied on for decades?

Traditional identity and access management was built around people and predictable software. A user authenticated, requested access, and performed specific actions in defined systems. We understand the human identity lifecycle clearly in this context, including how to provision access. AI agents introduce a different lifecycle. They can be created quickly, act continuously, adapt across workflows, and interact with multiple applications, APIs, and data sources with far more autonomy.

The core identity principles still apply: every identity needs authentication, authorization, governance, and accountability. The velocity is very different. Instead of a few enterprise applications built by centralized teams, organizations may have thousands of agents created by business users. In fact, 92% of CISOs lack full visibility into AI identities, and 95% doubt they could detect or contain misuse if it happened. Identity security has to evolve so that ownership, access controls, approvals, and monitoring are automatically established as agents are created.

Many organizations still use permissions and access controls that were designed for human employees and conventional software applications. Why are those approaches insufficient when dealing with AI agents that can reason, adapt, and execute actions independently?

Traditional authorization usually asks, “Is this identity allowed to perform this action?” That is no longer enough for AI agents because an agent may technically have permission to do something, yet still act outside the user’s intent.

For example, if a user asks an agent to “clean up” a database, the agent might interpret that as deleting records. It may be authorized to delete those records, but that may not be what the user meant. With agents, organizations need to evaluate not only whether an action is allowed, but whether it aligns with the original plan and intent behind the task.

Saviynt recently introduced Intent-Aware Runtime Authorization™. What exactly does “intent-aware” authorization mean, and how does it differ from traditional authentication and authorization models?

Intent-aware authorization means evaluating whether an agent’s actions align with what the human intended it to do. Every agent is created or dispatched to accomplish a task, and that task creates context for what the agent should and should not do.

Traditional authorization makes a binary decision: is this allowed or not? Intent-aware authorization adds another layer by asking whether the action makes sense in the context of the original task. If an agent were asked to archive old data but starts deleting active records, that may be technically permitted, but it is not aligned with the user’s intent. The goal of intent-aware runtime authorization is to govern agent actions in real time based on task context, policy, and intended outcomes.

How can enterprises determine whether an AI agent should be allowed to perform a specific action even when it technically has permission to access the underlying system?

It starts with planning. Before an agent takes action, organizations need to define what it is supposed to accomplish, which systems it should access, which guardrails apply, and what policies it must follow. That plan becomes the baseline for evaluating the agent’s behavior.

The key question is not just whether the agent has access, but whether its action is consistent with the task it was created to perform. Many existing applications and APIs were not built to enforce that level of context, which is why a gateway model is important. Agent Access Gateway provides an added control layer to evaluate plans versus actions, and help manage overentitlement or gaps in legacy systems.

One of the biggest concerns around AI agents is that they can act across multiple applications and data sources in seconds. What new risks does this create for security teams, and which risks are currently being underestimated?

AI agents do not always create entirely new risks. In many cases, they increase the likelihood that existing risks will be exploited. If an organization already has overpermissioned accounts, weak controls, or systems that rely on obscurity, agents can find and act on those weaknesses much faster than humans, scaling risk.

I sometimes compare it to leaving a back door unlocked. The vulnerability already exists, but the likelihood of someone finding it may be low. Now imagine an agent checking every back door across thousands of environments in seconds. That is why organizations need to get the basics right: know what agents exist, understand what they can access, limit unnecessary permissions, and govern their activity continuously.

As enterprises deploy multiple AI agents across departments, how should organizations govern who can interact with those agents and what information or systems those agents can access?

A helpful way to think about an AI agent is as an application. Enterprises already know how to govern applications: who owns them, who can use them, what systems they connect to, what data they can access, and how that access is reviewed. Those same questions apply to agents.

The difference is that agents can be created in minutes by business users, not over months by large development teams and only 5% of CISOs feel confident they could contain a compromised AI agent. Governance needs to be built into the creation process so ownership, access, approvals, entitlements, and periodic reviews are established from the beginning rather than added after the agent is already operating.

Shadow AI has become a growing concern as employees adopt AI tools outside official IT oversight. How do AI agents amplify this challenge, and what practical steps can organizations take to regain visibility and control?

AI agents amplify Shadow AI because they are easy to create and can act across systems. That does not mean organizations should try to stop people from using them. The productivity gains are real, and businesses will pursue them.

The practical answer is visibility and guardrails. Organizations need to know what agents exist, who owns them, what they are intended to do, and what systems or data they can access. Each agent should have an identity, defined permissions, and ongoing monitoring. Many enterprises are still in fairly early stages of experimentation, so there is still time to build governance before agent adoption reaches a much larger scale. But that time is quickly running out.

Looking ahead, do you believe every AI agent will eventually require its own digital identity, governance framework, and audit trail similar to human employees? What does that future model look like?

Every AI agent should unequivocally have its own identity so organizations know what it is, who owns it, what it can access, and what it is doing. Human identities and non-human identities, including AI agents, must all be governed through a unified identity security model.

The concepts are familiar: authentication, authorization, ownership, entitlements, activity tracking, and governance. What changes is scale. Enterprises may soon create hundreds of agents per week, eventually leading to tens of thousands or even millions of agentic applications. The future is about applying identity security fundamentals to every type of identity at a much greater speed and scale. Now is the time to start implementation.

As enterprises race to deploy agentic AI, what are the most common mistakes you see organizations making today, and what should security and technology leaders prioritize over the next 12 to 24 months to ensure AI agents remain both productive and trustworthy?

It would be a major oversight to leave security in this age to chance. The productivity gains related to agentic AI are real, and organizations need to capture them. The biggest risk is that security and governance do not keep pace.

Over the next 12 to 24 months, leaders should focus on visibility and governance fundamentals: maintain an inventory of agents, assign ownership, give each agent an identity, understand what it can access, and make governance part of the creation process. This is an opportunity to build security into a major platform shift from the outset instead of trying to retrofit it later.

Thank you for the great interview, readers who wish to learn more should visit Saviynt.