IAM in the Age of AI Agents: How to Navigate It Thoughtfully

The identity security market is undergoing a meaningful structural shift. AI agents, autonomous software programs that execute multi-step workflows across systems without human interaction, are becoming a standard fixture of enterprise IT.

According to Gartner, 33% of enterprise software applications will include agentic AI by 2028, up from less than 1% in 2024. That trajectory represents a significant change in the composition of enterprise identity populations, and it carries real implications for how organizations govern access.

Traditional Identity and Access Management platforms were designed for a well-understood problem: governing human workforce identities through structured lifecycle processes. SailPoint, Saviynt, and their peers built sophisticated platforms for exactly that model, and those platforms continue to serve that purpose well.

What is changing is that AI agents introduce a class of identity that operates outside the assumptions those platforms were built around -- and organizations that recognize this distinction early will be better positioned to address it deliberately rather than reactively.

This paper reflects the IDMWORKS advisory team's perspective on where identity security is heading, what we are observing across client engagements, and how organizations can approach AI agent identity governance with clarity rather than urgency.

Three Observations

• The governance frameworks built for human workforce identity do not map cleanly to AI agent identity, not because IGA platforms are failing, but because agents represent a fundamentally different kind of identity problem.
• The detection and response capabilities that security operations platforms have always provided are increasingly relevant to identity governance in an agentic environment, not as a replacement for IGA, but as a necessary complement to it.
• The organizations that navigate this well will do so by building a deliberate architecture that connects governance and detection, starting with definition and discovery rather than vendor selection.

Our Position

AI agent identity is not a future problem waiting to be solved. It’s an architectural gap that is widening now. Traditional IGA governance and security operations detection have always been distinct disciplines.

What’s new is AI agents make connecting them a business-critical requirement. Organizations that address this deliberately, starting with definition and visibility rather than vendor selection, will be better positioned than those that wait for the market to settle.

A Market at an Inflection Point: What AI Agents Actually Are

An AI agent is a software program that uses large language models to autonomously plan and execute multi-step tasks. Unlike traditional software, agents are not triggered by a user clicking a button. They are given a goal and pursue it by calling APIs, querying databases, sending communications, and modifying configurations without human involvement at each step. 

Microsoft, Google, Salesforce, ServiceNow, and most other major enterprise software vendors are embedding these capabilities directly into their platforms, meaning agents will increasingly be provisioned across typical enterprise environments not as a deliberate IT initiative, but as a byproduct of the software organizations already use.

Each of those agents is an identity. Each requires access rights. And each represents a potential attack surface that most IAM programs are not yet designed to address.

The Definition Problem

One of the most consistent observations we carry from client engagements is that organizations are attempting to govern AI agents before they have established a shared internal definition of what an agent is. The question of what qualifies as an AI agent in a given environment, versus a traditional automation, a service account, or an RPA process, is not a trivial one, and the answer varies meaningfully by organization.

Governance frameworks, credential policies, and architectural decisions all depend on that definitional foundation being in place first. Discovery and definition are not preliminary steps to be skipped in favor of vendor evaluation; they are the work.

What the Data Shows

The research picture on AI agent identity is still forming, but several findings from credible sources are worth noting:

33% of enterprise software applications will include agentic AI by 2028 (Gartner).

44% of organizations use static API keys to authenticate AI agents (CSA/Strata, 2026).

18% of security leaders are highly confident their IAM can manage agent identities (CSA/Strata, 2026).

It's worth noting that some statistics circulating in this space originate from vendors with a direct commercial interest in the problem they are measuring. We have been deliberate in anchoring to research from independent analyst firms and cross-organizational bodies where possible, and we would encourage the same discernment when evaluating vendor-produced data in this category.

What RSAC 2026 Confirmed

At RSA Conference 2026, securing AI agents as first-class corporate identities was among the dominant themes, with independent findings from IBM, Microsoft, and Ponemon Institute pointing in the same direction. The consensus among enterprise security practitioners: traditional IAM frameworks were not designed for the agent identity problem, and the industry is actively working to define what adequate governance looks like.

IBM Think, April 2026: "Traditional IAM was built around human users: login events, sessions, and access reviews. AI agents don't work that way. They are non-human, always-on, and operate across systems without a UI. The security and access layer underneath becomes the critical control point."

Traditional IAM Was Not Built for This: How Legacy IGA Platforms Work

IGA platforms like SailPoint and Saviynt are built around a well-defined architecture: a centralized database of identities, roles, entitlements, and policies, supported by provisioning and de-provisioning workflows, access certification campaigns, and audit reporting.

This architecture was designed for human workforce identity management, and it works well when identities are humans with known roles and business context, access events are discrete sessions with defined start and end points, provisioning follows HR-driven lifecycle events, and governance happens through periodic human review of access rights.

These are reasonable and durable assumptions for the workforce identity problem. They are the wrong assumptions for AI agents.

The architectural gap this table describes is not a criticism of IGA vendors. It's a reflection of the fact that these platforms were purpose-built for a different identity class. SailPoint and Saviynt are actively investing in agentic capabilities.

SailPoint's Harbor Pilot and Saviynt's Identity Control Plane for AI Agents represent meaningful steps in that direction, and we expect that investment to mature. The more useful framing is that AI agent identity introduces requirements, particularly around real-time detection and behavioral monitoring, that have historically belonged to a different part of the security stack, and that the industry is now working to connect those two disciplines more explicitly.

The Credential Problem

A 2026 joint survey by the Cloud Security Alliance and Strata Identity of 285 IT and security professionals found that 44% of organizations are authenticating AI agents with static API keys, 43% with username and password combinations, and 35% with shared service accounts.

These are credential patterns that IAM programs have spent years working to eliminate from human identity environments, now reappearing at scale in an identity category that most governance frameworks do not yet cover. When a single agent credential is compromised, the blast radius extends to everything that agent can reach, which is often far broader than any individual human account.

The Identity Debt Problem

Nearly 80% of organizations deploying autonomous AI cannot report in real time what those systems are doing or who is accountable for them (CSA/Strata, 2026). AI agents are being deployed into this governance gap by default -- and the debt accumulates silently until it becomes a remediation problem.

Where Identity Control Is Moving Next: The Role of Security Operations Platforms

Vendors like CrowdStrike, SilverFort, and Fortinet are increasingly relevant to the identity security conversation -- not because they are displacing IGA, but because AI agent identity looks more like a security operations problem than a governance problem. Security operations has always been in the business of monitoring behavioral signals, correlating event telemetry, and detecting anomalous activity in real time.

That discipline is not new. What is new is that AI agents create an identity challenge that maps naturally to those existing capabilities in ways that traditional access governance does not.

The governance layer answers the question of who has access to what, managed through workflows and periodic review. The detection layer answers the question of what is actually happening right now, across every surface. IGA was designed for the former.

Security platforms were designed for the latter. AI agents -- autonomous, always-on, operating across multiple systems simultaneously -- require both, and they require them to be connected. The organizations best positioned to manage this are those that approach it as an architecture question rather than a vendor selection question.

How the Market Is Responding

The identity security market is responding to this shift from multiple directions simultaneously. IGA vendors are extending their platforms toward agentic governance.

Security vendors are formalizing their identity capabilities. Analyst firms including Gartner are developing new category frameworks -- such as Identity Visibility and Intelligence Platforms (IVIP) -- to describe the unified intelligence layer that connects IAM, SIEM, and endpoint telemetry.

The market has not yet settled on a definitive architecture for AI agent identity governance, and that ambiguity is appropriate given how early the problem is. What is clear is that organizations waiting for a single-vendor solution to emerge before they begin addressing the problem will face a more complex remediation effort later.

The Strategic Implication

The value in identity security is shifting toward a model that combines governance (who has access to what) with detection and enforcement (what is happening, in real time, across every surface). The enterprises that navigate this well will not choose one or the other -- they will build an architecture that connects both.

5 Implications for Enterprise IAM Programs

We advise clients to approach AI agent identity governance as a staged discipline rather than an urgent overhaul. The following sequence reflects the order in which the underlying problems need to be solved.

A Framework for Moving Forward

1. Establish Agent Identity Standards Before Scale: Define how AI agents will be created, credentialed, and governed in your environment before they proliferate. This includes naming conventions, credential rotation policies, least-privilege scoping, and decommissioning workflows. The cost of establishing these standards proactively is substantially lower than imposing them retroactively, and that cost differential only grows as agent populations scale.

2. Inventory Your AI Agent Population: Most organizations do not have a clear picture of the service accounts, API keys, and non-human identities currently operating outside IAM visibility. A discovery exercise (scanning for non-human identities not attached to a human owner) is the foundational operational step. You cannot govern what you cannot see.

3. Extend Identity Visibility Beyond the IGA Platform: Evaluate whether your current architecture provides adequate coverage for identity activity that occurs outside traditional IAM governance. Gartner's emerging category of Identity Visibility and Intelligence Platforms (IVIP) -- unified identity intelligence spanning IAM, SIEM, and endpoint telemetry -- is worth assessing alongside existing IGA investments rather than in isolation from them.

4. Evaluate Security-First Identity Vendors: If your IAM architecture is built primarily around SailPoint or Saviynt, assess whether your detection and response capabilities for non-human identities are adequate for an agentic environment. Platforms like CrowdStrike and SilverFort are worth evaluating as complements to your existing governance layer -- and in some cases, as the primary identity security layer for agentic workloads specifically.

5. Assess Your IGA Vendor's Roadmap Honestly: SailPoint and Saviynt are investing in agentic capabilities. Understand what is available today, what is on the roadmap, and what the architectural constraints are. For organizations with complex, heterogeneous environments, a multi-vendor approach (IGA for workforce governance alongside a security platform for agent detection and response) is often the most practical path forward.

The IDMWORKS Perspective

IDMWORKS has been advising enterprises on identity and access management for over a decade, across healthcare, financial services, energy, manufacturing, and higher education. We have worked alongside the platforms that govern workforce identity, the security vendors that detect and respond to identity threats, and the enterprise teams responsible for connecting both.

The transition to agentic AI is the most significant architectural shift we have seen in this space, and our perspective is shaped by the advisory and implementation work we do every day across that full spectrum.

Our view:

• Traditional IGA remains essential for workforce identity governance. SailPoint and Saviynt are not going away. The human identity governance problem is not solved.
• AI agent identity requires a complementary architecture that adds detection and enforcement capabilities to the governance foundation. The two disciplines are not in competition -- they address different parts of the same problem.
• Security vendors are building the unified identity layer. CrowdStrike, SilverFort, and Fortinet are formalizing identity threat detection, endpoint security, and network control in ways that matter for agent governance.
• The enterprises that navigate this transition well will not choose a single vendor. They will build an architecture that uses IGA for workforce governance, security platforms for detection and response, and a unified visibility layer to connect them.

IDMWORKS helps organizations navigate exactly this transition. Our work spans assessment and strategy, IGA implementation, security vendor integration, and agentic AI readiness.

We bring an independent, architecture-first perspective across the full vendor landscape -- SailPoint, Saviynt, CrowdStrike, SilverFort, Fortinet, Microsoft Entra -- because that independence is what allows us to give clients guidance that serves their environment rather than any single vendor's roadmap.

What This Means for You

AI agent identity is a present and growing challenge, and the organizations addressing it now are doing so from a position of advantage. The identity infrastructure most enterprises have built was designed for a different kind of identity population -- and extending it thoughtfully to cover AI agents is achievable, provided organizations start with definition and discovery rather than jumping to architecture or procurement decisions.

The market is still developing the frameworks, platforms, and standards that will eventually make AI agent governance routine. Traditional IGA platforms are adapting. Security vendors are formalizing their identity capabilities. The organizations that recognize this shift and build toward it intentionally in 2026 will be meaningfully better positioned than those that wait for the landscape to settle.

IDMWORKS serves as the bridge between where identity governance has been and where it needs to go, bringing the advisory depth, implementation experience, and vendor-neutral perspective that this transition requires.

Are You Ready for the Agentic Era?

IDMWORKS offers a focused AI Agent Identity Readiness Assessment; a structured engagement that inventories your non-human identity population, evaluates your current governance gaps, and provides a prioritized roadmap. Reach out to us today for next steps.