Securing Identities For The Agentic AI Landscape

Forbes

03-07-2025

Art Gilliland, CEO at Delinea.
The rise of agentic AI systems—autonomous, self-directed entities capable of making decisions and taking actions—marks a significant shift in the digital landscape. These aren't just smarter tools executing pre-scripted tasks; they're intelligent agents capable of reasoning, learning, adapting and operating independently across complex environments. As organizations begin deploying these systems at scale, the implications for identity security are both profound and urgent.
From Assistive To Agentic: The AI Evolution
We've long worked alongside assistive AI and think about search engines, chatbots and workflow automation. These systems' enhanced capabilities, however, required explicit instructions. Agentic AI changes the game. It doesn't just respond—it initiates. It can proactively launch tasks, coordinate with other agents, secure resources and continuously optimize toward goals, often with little or no human oversight.
This shift is unlocking powerful new use cases across industries. In healthcare, agentic systems optimize patient flow by autonomously managing bed assignments, staffing levels and hospital transfers in real time. In financial services, they execute dynamics, reallocate portfolios and manage trading strategies and risk exposures independently. In government, agentic AI is reshaping emergency response—deploying resources, rerouting supply chains and prioritizing services based on live data. Across sectors, AI is no longer just a support tool; it's a decision-maker.
This evolution introduces a new identity type: autonomous agents requiring access, permissions and authority within enterprise systems. Traditional models—built for humans and static accounts—can't support this. We need new frameworks built for AI-native environments.
New Threats, New Surfaces
As agentic AI becomes more capable, threat actors are adapting just as fast. Identity is quickly becoming both the first line of defense and the first target.
Malicious actors are already exploiting synthetic identities, with AI entities posing as legitimate agents to infiltrate systems. Some agents, through repeated learning loops, are discovering ways to exploit access gaps or unintentionally escalate privileges. Insider threats are evolving too, with employees deploying unsanctioned AI agents to bypass controls, exfiltrate data or automate prohibited tasks under the radar. Even data inputs are a risk—manipulated datasets and poisoned outputs can deceive agentic systems into making harmful decisions.
Agentic AI demands agentic security. Organizations must rethink identity, building adaptive, risk-aware systems that treat every action, whether human or machine-driven, as a decision point that needs to be verified, validated and secured.
Rethinking Identity In The Age Of Autonomous Agents
To securely enable agentic AI, organizations must treat these entities as privileged employees, contractors or partners. To do that demands significant changes in how identity and access management (IAM) is architected:
Lifecycle Management: Like human workers, AI agents must have well-defined onboarding, role assignments, periodic access reviews and offboarding processes. Identity governance must accommodate dynamic, non-human actors whose roles and scopes of responsibility evolve over time.
Context-Aware Authorization: Static access policies are inadequate. Agentic AI may pivot across tasks, applications and datasets within seconds. Access decisions must be adaptive, based on real-time context—task type, risk level, location, behavioral history—and capable of adjusting dynamically.
Non-Repudiation And Accountability: Every action taken by an AI agent must be traceable to maintain trust, ensure compliance and enable forensic analysis if needed. Robust logging, cryptographic verification, immutable audit trails and transparency are essential.
Zero Standing Privilege (ZSP): AI agents should never retain persistent access to sensitive resources. Just-in-time (JIT) privilege elevation—where access is granted only when needed and automatically revoked afterward—is key to minimizing the blast radius of any potential compromise.
Building Trust Into Intelligent Autonomy
Securing agentic AI requires more than just layering on new security tools—it demands a fundamental shift in our trust models. As machines gain the autonomy to make decisions, we must redefine how authority is established, intent is verified and boundaries are enforced. This evolution calls for proactive, cross-functional collaboration across the organization.
At Delinea, we see identity as the core of securing AI. That means identity and security teams must build adaptive access controls and policy-based automation into every phase of the AI lifecycle. AI/ML practitioners must embed least privilege and explainability into models from the start. Governance and compliance leaders must ensure policies scale with machine identities, while legal and risk teams revisit accountability in a world where machines act on behalf of humans. By integrating identity, intent verification and privilege management directly into AI systems, organizations can secure autonomy without sacrificing control.
5 Steps Companies Should Take To Manage Agentic AI
To secure the evolving agentic AI landscape, organizations should follow a structured approach:
1. Discover And Classify AI Identities: Use automated tools to inventory AI agents—scripts, bots and autonomous models, across hybrid and multi-cloud environments. Classify by sensitivity, privileges and business impact to align with intelligent privilege management.
2. Define Roles And Guardrails: Set clear operational boundaries for each AI identity. Use policy-based access to tie privileges to specific tasks, keeping actions aligned with business intent and risk tolerance.
3. Enforce Least Privilege, Just-In-Time Access: Replace standing privileges with just-in-time access. Grant AI agents only what's needed when it's needed—then revoke it automatically to reduce risk.
4. Authenticate And Authorize By Intent: Require strong, verifiable identities for AI-to-system and AI-to-AI interactions. Go beyond identity to validate intent, ensuring actions match approved use cases.
5. Monitor, Detect And Continuously Improve: Continuously monitor AI behavior to detect anomalies and misuse. Log actions with cryptographic integrity, enforce encryption and regularly test workflows to harden identity and access controls.
The Path Forward: Building Trust In Agentic AI
The agentic AI era isn't approaching; it is already here. From code co-pilots that autonomously write and deploy software to intelligent orchestrators managing entire workflows, AI agents are rapidly embedding themselves in critical systems.
Organizations that embed identity-first principles into their AI strategies will be best positioned to innovate securely, maintain stakeholder trust and mitigate emerging cyber risks. Those who delay risk falling behind, not only technologically, but in defending against adversaries moving as fast as the machines themselves.
Let's make sure the future is secure.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?