The agentic enterprise is moving rapidly from concept to reality, says John Cannava, CIO of the platform Ping Identity.
As artificial intelligence (AI) transitions from passive assistant to active operational agent – executing workflows and making decisions alongside employees – a critical question arises: How can organisations responsibly harness AI to strengthen trust without sacrificing human productivity or user experience?
For decades, identity architecture was engineered around human users and static, predictable systems. In the age of agentic AI, those legacy models fall short. AI agents operate continuously across corporate environments, yet many enterprises lack clear visibility into which agents exist, what endpoints they connect to, what data they access, and who authorised their deployment in the first place. This lack of visibility creates severe operational and security vulnerabilities, leaving IT and security leaders struggling to balance rapid innovation with effective risk management.
Addressing this shift requires a cross-functional approach shared between IT and security leadership. Rather than treating AI governance as a reactive security fix or an add on feature, forward-thinking enterprises must build AI capabilities directly into the core of their identity strategy. Systems, data access, and user experience must be re-evaluated to support both the human workforce and autonomous digital actors. To operate safely at scale, identity must evolve from a static, point-in-time gatekeeper into a continuous control plane capable of evaluating every request in real time against policy and dynamic risk levels.
Enterprise-wide agent discovery
AI agents are entering production faster than traditional governance models can adapt. Before security teams can govern these non-human actors, they must first identify them. Organisations require a unified, real-time inventory of every AI agent across both workforce and customer-facing applications – ranging from enterprise copilots and third-party tools to emerging personal agents interacting with corporate APIs.
Continuous agent discovery provides the baseline for modern security and operational governance. It enables security leaders to audit compliance, quantify data exposure, and transition AI governance from a reactive response to a proactive strategy. Identity platforms must naturally expand their perimeter to continuously catalog users, devices, and autonomous digital actors simultaneously. Once cataloged, these entities require dedicated authorisation controls.
Explicit delegation over inherited privilege
For CIOs, the spread of AI tools creates a familiar challenge: Balancing innovation at the edge with centralised visibility and control. Without centralised control and limiting third-party AI tools to enterprise data and systems, organisations face heightened risk of data leaks, inconsistent security enforcement, and regulatory non-compliance.
At the root of this challenge lies over-privilege. Assigning broad directory roles or default user-lever access to autonomous agents introduces unnecessary risk, exponentially expanding the potential blast radius should an agent or credential be compromised. The solution is a deliberate shift toward explicit delegation – granting machine identities narrowly scoped, purpose-built permissions.
Security leaders must evaluate key access parameters: Who is authorised to connect AI tools to enterprise repositories, under what specific conditions, with what scope, and for how long? Authorisation can no longer be statically assigned at login; it must be contextual and evaluated continuously at runtime.
Implementing policy-driven controls at the identity layer – such as strong authentication for machine identities, constraints on high-risk actions, and explicit step-up approvals for sensitive data – ensures agents operate strictly within their intended boundaries. Identity must move beyond validating who or what an agent is to determining precisely what it is permitted to execute in that exact moment.
Behaviour monitoring for continuous trust
Unlike static software, AI systems are dynamic. Underlying models drift, system prompts evolve, connected data sources expand, and agent behaviours adapt over time. Consequently, trust cannot be granted via a one-time setup; it must be continuously evaluated.
Effective governance demands ongoing behaviour monitoring and runtime validation. This includes simulating adversarial misuse, monitoring for abnormal activity patterns, and detecting unexpected privilege escalations or behavioural drift. By adopting these practices, identity platforms evolve from rigid policy engines into dynamic trust evaluators – securing machine workflows behind the scenes so human employees can innovate without friction or risk.
Balancing innovation and security
Restricting AI adoption through overly rigid controls will only slow innovation, frustrate employees, and push adoption into the shadows, which can all create even greater risk. Itโs a delicate balance. ย The goal is to enable safe, sustainable, and scalable AI adoption. Achieving this requires a modern identity architecture that functions as a single control plane for humans and autonomous agents alike – one where authorisation is contextual, privileges are tightly scoped, and every interaction is governed in real time.
Trust must be continuously verified, never assumed. For CIOs and CISOs, embedding AI into the core identity strategy provides the foundation for an enterprise AI ecosystem that is both highly secure and operationally resilient – allowing automation, trust, and the human experience to advance together.
