Identity , AI/ML Laying the groundwork: A practical path to identity security for AI agents May 27, 2026 Share By Paul Wagenseil Credit: Adobe Stock Images AI agents, ranging from customer-facing assistants to autonomous digital workers, are becoming commonplace in enterprise environments. Yet unlike traditional applications, AI agents can reason, make decisions, and act independently. As a Ping Identity executive brief notes, agentic AI introduces "a new, novel class of non-human identity" that challenges traditional identity and governance models. Because of this shift, organizations must modernize identity-security systems originally built for human and machine accounts that used long-lived, static credentials. Enterprises must adopt identity frameworks that provide visibility, just-in-time runtime authorization, delegated access, and human management. The importance of visibility and human oversight Adapting identity security for AI agents begins with visibility. As the old saying goes, you can't manage what you can't see. Accordingly, security teams must discover, classify, and continuously monitor every AI agent in the environment. Agents that mimic human behavior through desktop browsers or graphical interfaces also require detection and tagging. Each AI agent, no matter its manner of interface, should be given a distinct identity, linked to an owner or custodian, and governed through lifecycle controls similar to those of workforce identities. Human oversight is equally important because some actions are too sensitive to execute autonomously. In a recent overview of upcoming identity trends , Ping Identity recommends human-in-the-loop (HITL) controls for high-risk transactions such as financial approvals, changes to privileged access, or sensitive data operations. Modern identity systems must support step-up authentication, out-of-band approvals, and continuous auditing . Keeping humans in the loop means keeping accountability, maintaining compliance, and preventing agents from exceeding their authority. How to authenticate and authorize AI agents Modern identity architectures should treat agents as first-class digital identities with scoped permissions, runtime authorization, and policy-driven controls. In an online guide to managing AI , Ping Identity recommends using OAuth 2.0 delegation, Dynamic Client Registration (DCR), assertion grants, and mutual TLS (mTLS) to authenticate and authorize agent actions securely. Short-lived access tokens and just-in-time (JIT) entitlements are critical to enforcing the principle of least privilege. Instead of granting permanent access to AI agents, organizations should evaluate permissions continuously based on context, task intent, and risk signals. Organizations must also facilitate secure interoperability between agents, APIs, and external tools, which can be provided by standardized frameworks such as the Model Context Protocol (MCP) and the Agent-to-Agent (A2A) protocol, which also can help enforce in-house policies. Why humans should never share credentials with AI agents In a recent explainer , Ping Identity warns that one of the most dangerous practices in agentic AI environments is credential sharing, as it destroys auditability and creates significant risks around impersonation, fraud, and unauthorized access. Gartner predicts that by 2028, organizations that let humans share credentials with AI agents will triple their rates of account takeover and fraud. Instead, organizations should use authenticated delegation, in which agents receive scoped, temporary tokens that define exactly what they can do and on whose behalf they can act. This preserves accountability while preventing agents from inheriting unrestricted human privileges. Delegated authorization also enables full audit trails. Security teams can trace every action back to the specific agent identity, delegated authority, and policy context that permitted the transaction. Combined with runtime monitoring and automated revocation controls, this reduces the blast radius of compromised or misconfigured agents. "Being AI-ready is no longer aspirational — it's operational," says a Ping Identity guide . "This means having the governance, security, and infrastructure to safely scale AI agents across customer and workforce environments." Paul Wagenseil Paul Wagenseil is a custom content strategist for CyberRisk Alliance, leading creation of content developed from CRA research and aligned to the most critical topics of interest for the cybersecurity community. He previously held editor roles focused on the security market at Tom’s Guide, Laptop Magazine, TechNewsDaily.com and SecurityNewsDaily.com. Related Identity A new identity class: Why AI agents require runtime control Paul Wagenseil May 27, 2026 Because AI agents are neither human nor traditional machines, they must be treated as a third class of identity. Identity 7 identity security best practices for the Agentic AI era Neal Goldman May 22, 2026 Here’s how to harden the teams identity security to defend against the rising tide of AI agents. Identity How identity became the new security battleground Craig Birch May 21, 2026 Exploits happen in minutes in the AI era – so 43 days to fully remediate just doesn’t cut it today. Related Events Cybercast IAM for MSSPs: Real-World Deployments On-Demand Event Cybercast Privilege risk is in the lifecycle: A CISO discussion on modernizing identity control On-Demand Event Cybercast The industrialization of identity compromise On-Demand Event Get daily email updates SC Media's daily must-read of the most current and pressing daily news Business Email By clicking the Subscribe button below, you agree to SC Media Terms of Use and Privacy Policy . Subscribe Related Terms Basic Authentication Biometrics Certificate-Based Authentication Challenge-Handshake Authentication Protocol (CHAP) Digest Authentication Digital Certificate Discretionary Access Control (DAC) You can skip this ad in 5 seconds