I went into Identiverse 2026 with a specific curiosity about where the intersection of AI and identity security was heading, but I came out realizing something far more urgent. Most organizations are dangerously unprepared to govern the identities they are about to unleash.
We’ve all read the headlines tracking the rapid rise of enterprise AI. AI agents are no longer just passive, deterministic chatbots; they are autonomous, goal-directed entities that can reason, execute complex multi-step tasks without hand-holding, and make data-driven decisions at a speed and scale no human team could match. Industry analysts validate this massive shift, predicting that up to 40% of enterprise applications will integrate task-specific AI agents by the end of the year. The productivity pitch practically writes itself: compounding operational efficiency, streamlined workflows, and significantly lower costs.
But as a product leader, my mind kept returning to an existential structural blind spot. How will enterprise security teams evolve one of the most established controls in identity governance: certification?
The stark reality is that traditional governance frameworks were architected for a human-centric world. In a modern enterprise network, non-human identities (NHIs) — including service accounts, tokens, APIs, RPA bots, and autonomous agents — now vastly outnumber human employees by an alarming ratio of 82:1. When you introduce autonomous, machine-speed execution into that lopsided estate, our traditional identity safety checks begin to fracture.
The core paradigm shift: It’s no longer about assigned access
For years, access certification has served a simple, relatively predictable purpose. People accumulate permissions over time, organizations review those entitlements periodically (usually on a quarterly or annual cycle), and managers attest that the access still makes business sense. It’s never been a perfect process, but it has generally functioned because human identity patterns are predictable and slow-moving. Human employees change roles or switch departments, but their fundamental responsibilities and access patterns don’t completely rewrite themselves every few days.
Agentic AI completely shatters that core assumption.
Consider the practical scenario of an AI procurement agent. When first deployed, its purpose is narrowly defined and clear — it creates purchase requests, gathers vendor information, and routes internal approvals. On paper, its initial access registration and certification are perfectly accurate and low-risk.
However, because these systems are dynamic and contextual, the environment around that agent changes rapidly. Weeks or even days later, the agent might connect to new tools via Model Context Protocol (MCP) frameworks, interact with entirely different sets of automated agents, execute tasks on behalf of shifting end-users, or weave itself into workflows that didn't even exist when it was originally approved.
Here lies the problem: nothing about that original, point-in-time certification was wrong. Yet, the agent’s effective authority has evolved entirely. What an autonomous agent can actually accomplish in a live production environment is no longer restricted to what was explicitly assigned on a static configuration sheet. Instead, its true privilege is a fluid, real-time combination of its own toolsets, the permissions of the human who registered it, the entitlements of the user it acts for, and the target applications it can reach through those dynamic relationships.
Traditional identity governance was engineered to validate assigned access. In an agentic world, the real security risk stems from what an identity can actually do in its current, shifting context.
Certification needs to evolve
The limits of point-in-time access reviews boil down to velocity and context. Traditional certification operates on the assumption that identity risk changes slowly enough to be caught during a scheduled batch review campaign.
For autonomous agents, that latency is an active liability. An agent can spawn a new connection, chain a tool call, or inherit an unreviewed credential silently and instantaneously. Waiting for a quarterly compliance review to flag that exposure is an obsolete defense. By the time the next certification campaign runs, an over-privileged or rogue agent could have already triggered a severe security event.
The breakdown isn't happening because security teams are running their compliance campaigns poorly. It's happening because scheduled reviews are fundamentally disconnected from the precise moment that trust changes.
Historically, identity compliance has asked a simple question: Does this identity still need this access? In an agentic ecosystem, we must ask a much more critical question: Should this identity still be trusted to exercise the authority it currently holds, in this specific context, for this precise action, and across this complex web of relationships?
Access tells us what an identity can do on paper. Trust determines whether it should continue doing it in reality. Navigating that distinction will be one of the primary challenges in identity governance.
Shifting to Continuous Certification of Trust
I am increasingly convinced that the industry must transition toward Continuous Certification of Trust as a foundational governance principle for non-human and agentic identities.
To be clear, periodic certifications won't vanish overnight. Scheduled reviews will always play a role in baseline regulatory compliance, oversight, and organizational accountability. But as a standalone defense, point-in-time audits are no longer enough.
As agents continuously acquire new capabilities, tap into disparate systems, and execute multi-step chains, governance must operate in lockstep with runtime activity. Instead of relying solely on scheduled reviews, organizations require an automated, context-aware mechanism to continuously evaluate whether a digital identity remains strictly within the boundaries of its originally sanctioned trust posture.
Operationalizing the future
The strategic direction we need to take is obvious, and operationalizing this model is a challenge my Saviynt team and I are actively working to solve.
Transitioning to a continuous trust framework surfaces tough, pragmatic questions for identity architects:
- Who or what actually owns the remediation decision when an agent's trust score drifts out of bounds? Is it a human manager, an automated policy engine, or a hybrid combination of both?
- How do we implement continuous verification without turning the process into another overwhelming source of alert fatigue?
These are not abstract design theories. They are the exact operational and identity-platform challenges that will determine whether this model works in practice. The conversation has evolved beyond the question of whether certification needs to change. Our focus should be on how organizations will effectively balance dynamic, continuous trust with enterprise security and usability.
To learn more about how Saviynt secures AI agents in today's dynamic environment, visit saviynt.com/products/identity-security-for-ai.
Back to Blog
.jpg)
