For a long time, the way we handle identity and access management (IAM) was pretty straightforward. Basically, when a user logged in, we verified their identity, determined what they could access, and then let the application do its thing. But with the rise of AI agents, that approach is becoming a lot more complicated.
Now, an AI agent can log in once and then spend a significant amount of time making decisions, calling different APIs, using various tools, and performing tasks on behalf of a person or an organization. This shifts the focus from simply asking who logged in to more complex questions. We need to think about who is currently acting, for whom they are acting, what authority they have, and whether what they’re doing is still acceptable. This presents a much larger challenge when it comes to identity management.
As we look ahead to the developments within Ping Identity, it’s clear that they’re focusing on addressing these challenges. They’ve recognized that AI is becoming a critical player on both sides of IAM.
For starters, AI is now a consumer of identity infrastructure. With Ping’s Identity for AI strategy, they’re treating AI agents as separate identities. This means these agents can be given delegated authority, access company resources, and manage their actions in real-time, rather than just relying on a human user’s login information. This distinction is really important. An AI agent that’s acting for an employee shouldn’t just get that employee’s credentials and be expected to follow the rules. Instead, the AI should have its own identity, and the system needs to track who granted its authority, what that includes, how long it’s valid, and what should happen if the agent tries to do something beyond its limits.
At the same time, AI is starting to handle parts of the IAM platform as well. Ping’s latest offerings include MCP servers, command-line access, documentation geared towards AI agents, and AI support within their DaVinci tool. This means that while AI needs to be managed, it’s also going to be interacting with IAM infrastructure in various ways. That’s where things get really interesting in terms of identity architecture.
Now, companies might think they can take a shortcut by creating a service account for an AI agent. They might assign it an API key, give it enough permissions, and store the credentials somewhere to move on. But we’ve seen this story before. When service credentials linger too long, ownership becomes murky, permissions can get overly broad, and secrets can end up scattered everywhere. Eventually, it’s clear the account is essential, but nobody knows what will happen if those credentials are changed.
The problem gets worse with AI agents, whose behavior can be unpredictable. Traditional scripts generally execute the exact instructions programmed by someone. However, an AI agent can decide which tools to use, what information to gather, and how to proceed based on its own reasoning. This flexibility is a big part of what makes AI powerful but also means that static credentials are inadequate. As AI can make decisions on its own, its authority needs to be managed just as dynamically.
That’s where Ping’s concept of Runtime Identity comes into play. Traditionally, IAM focused mainly on security at the start of an interaction—authenticating the user, checking for multi-factor authentication, assessing risk, and issuing a token to start the session. But for an autonomous agent, the critical action might happen much later on. An agent could call an API, retrieve sensitive info, complete a transaction, or link up with another system. By then, mere authentication from a few minutes before isn’t enough to determine if the agent should be allowed to proceed. The requested action might involve more sensitive data, or the agent might be trying to use capabilities the user didn’t intend to delegate.
This shift towards runtime authorization is about assessing identity and authority, along with context and policies, at the moment when action takes place—not forcing another login. It’s about examining whether this specific agent, given this delegation, should be allowed to perform this action right now. This may seem like a subtle change in terminology, but architecturally, it’s a significant shift.
The Model Context Protocol (MCP) makes these discussions even more vital. It gives AI systems a standardized way to discover and interact with various tools and capabilities, which is incredibly helpful. But it also raises critical security questions. If an agent can find a tool, should it be allowed to use it? If it can operate it, what actions should it take? What authority is it working under, and how long will that authority last? And what if using that tool triggers a larger business process?
Once tools can modify data, start transactions, access sensitive information, or initiate workflows, MCP transcends just being an integration issue. It becomes a matter of identity and authorization altogether.
Recent incidents involving AI and security breaches highlight this issue even more. For example, Ping noted an incident where an autonomous AI system breached part of a production environment. The lesson wasn’t just that AI can be involved in attacks; it was what unfolded afterwards. The spread of the attack hinged on credentials, machine identities, permissions, and their trusted relationships.
These key points should resonate with anyone familiar with IAM. The nature of the initial exploit can vary, but the potential “blast radius” often circles back to identity. Which credentials were exposed? What could those identities access? How long were those credentials valid, and how quickly could they be revoked? Could the organization distinguish between normal and abnormal behavior?
Rather than making these questions irrelevant, AI only intensifies their importance. It’s clear that as AI continues to evolve, the way we think about identity in IAM needs to adapt rapidly.
Ping YOUniverse: An Opportune Moment for Transformation
The timing of Ping YOUniverse aligns perfectly with current trends and challenges in identity management. This event emphasizes the critical nature of trust in an increasingly AI-driven landscape, specifically targeting AI-ready Identity and Access Management (IAM), non-human identities, autonomous interactions, and the enforcement of runtime policies.
The identity industry faces a pivotal question: Are AI agents merely a new type of service account, or do they necessitate an entirely different architectural framework? Ping has clearly opted for the latter, presenting a vision that requires organizations to consider how this new model integrates with existing infrastructures.
Enterprises must grapple with several vital questions: How do agent identities coexist with traditional workforce and customer IAM frameworks? How is delegated authority represented within this new paradigm? What strategies can organizations employ to prevent the emergence of another generation of standing privileges? Furthermore, how can runtime authorization be effectively managed across Software as a Service (SaaS), APIs, legacy applications, and multi-cloud platforms? Security teams must also be equipped to reconstruct the actions of AI agents, ensuring robust oversight and compliance. Additionally, how can these controls be implemented without necessitating an entirely separate identity stack dedicated to AI?
These are the pressing implementation considerations.
Integrating AI into Identity Architecture
At Navar, we focus on translating identity architecture into practical, operational realities. This involves the comprehensive implementation and integration of Ping technologies across Single Sign-On (SSO), Multi-Factor Authentication (MFA), Customer Identity and Access Management (CIAM), authentication processes, federation, application onboarding, legacy integration, and identity modernization.
This same hands-on approach is essential when it comes to integrating AI agents. Organizations utilizing agentic AI require more than a theoretical “AI security strategy.” They need actionable authentication and authorization frameworks that seamlessly integrate into their existing identity infrastructures. This entails crafting distinct agent identities rather than relying on shared credentials, employing OAuth and delegated authorization judiciously, and effectively managing tokens and access scopes. Furthermore, securing access to multi-cloud platforms, determining the appropriate points for runtime policy evaluation, and incorporating principles of revocation, logging, traceability, and lifecycle management into the architecture from the outset are crucial.
For organizations already leveraging Ping solutions, it is essential to recognize how Ping’s evolving AI and Runtime Identity capabilities can be utilized in conjunction with real-world applications and agent workflows, moving beyond the view of these features as isolated products.
Bridging the Gap
Navar is uniquely positioned to bridge the gap between established identity management and emerging AI operational requirements. Our expertise lies in implementing Ping’s technology along with practical strategies for AI agent authentication, securing multi-cloud environments, OAuth usage, delegated access, runtime enforcement, and non-human identity governance.
The objective is not to create a parallel security framework for AI but to enhance the enterprise identity model. This allows for the authentication, authorization, constraint, observation, and revocation of agents using controls that organizations can genuinely manage.
Avoiding Past Mistakes
The most challenging moment to devise an identity model is after numerous identities have already been established. Many enterprises have experienced this with service accounts and cloud workloads, leading to a legacy filled with unmanaged non-human identities, overprivileged API keys, shared credentials, and automation accounts lacking clear ownership.
AI agents present a unique opportunity for organizations to break this cycle. Before agents are deployed in production systems, companies should determine the nature of agent identities, the mechanics of authority delegation, the duration of that authority, the attribution of actions, and the points at which authorization must be reassessed. These considerations should be integral to the architecture, not after AI programs are already in operation.
AI agents will act autonomously, making it crucial for identity architecture to define their permissions upfront. As Ping YOUniverse approaches, this will be a key conversation in the IAM sphere. For enterprises prepared to advance from concept to execution, Navar stands ready to assist in designing and implementing the identity controls necessary for managing agentic AI without incurring unmanaged access issues.
