As organizations accelerate digital transformation and embed AI across business processes, applications and workflows, cybersecurity leaders face a growing challenge: the attack surface is expanding faster than traditional security models can adapt. Every new AI agent, cloud workload, application integration and automated workflow introduces additional pathways for compromise, increasing organizational exposure to cyber risk.

In this environment, organizations require a consistent security framework that enables risk-informed decisions while reducing implicit trust across the enterprise. Zero trust is gaining increased focus as the strategic paradigm for achieving this objective. While emerging technologies such as generative AI and autonomous AI agents introduce new security considerations, the core principles of zero trust remain unchanged. Organizations should align all security initiatives, including those designed to govern AI systems, with a common set of zero-trust principles to ensure consistent protection and governance.

Gartner recommends five foundational principles that cybersecurity leaders can use to guide security strategy, technology investments and program maturity.

1. Zero trust is a security paradigm, not a technology deployment

Gartner defines zero trust as a security paradigm that replaces implicit trust with explicit trust through continuous verification of identity, context and risk. Rather than relying on a single authentication event, access decisions should be continuously evaluated and validated.

Organizations must enforce explicit trust through mechanisms such as conditional access policies, multifactor authentication and continuous monitoring of identity, device and session risk signals.

2. Assume the presence of hostile actors

Traditional security programs often focus on preventing attackers from gaining access. Zero trust starts from a different assumption: attackers may already be inside the environment.

This mindset shifts the focus toward containment, resilience and rapid detection. Organizations should design controls based on the expectation that systems, accounts and even AI-enabled tools may eventually be compromised.

For AI environments, this means treating AI agents, external tools and retrieved content as potentially untrusted. Every agent action, tool invocation and external data source should be subject to validation, authorization and monitoring to reduce the risk of prompt injection, unauthorized actions and manipulation.

At the same time, organizations should continue strengthening foundational controls such as encryption, continuous monitoring and the reduction of unnecessary administrative privileges. These measures remain essential regardless of whether the workload is traditional or AI-enabled.

3. Establish identity as the foundation of access

Identity is at the heart of zero trust. Access decisions should be based on verified workload, user and device identities.

The rise of AI, however, expands the definition of identity. AI agents increasingly access systems, retrieve information and interact with applications on behalf of users. As a result, organizations must treat AI agents as distinct identities with their own authentication and authorization requirements. Every agent-to-system, agent-to-agent and agent-to-tool interaction should be authenticated before access is granted.

Without strong identity controls, organizations risk creating new blind spots and trust relationships that security teams cannot effectively govern.

4. Limit access to required functions

Zero trust is built on the principle of least privilege. Users, devices, applications and workloads which include AI agents should receive only the access necessary to perform their intended functions. Access should be granted explicitly, reviewed regularly and adjusted based on risk and business need.

For AI systems, this principle becomes especially important. AI agents do not require unrestricted access to enterprise resources to deliver value. Their access should be limited to the specific data, APIs and systems required to complete assigned tasks. Organizations should adopt a “deny by default, allow explicitly” approach and continuously review permissions to avoid privilege creep.

Limiting access reduces exposure and helps contain the impact of compromised accounts, systems and AI agents.

5. Apply risk-based adaptive access

The final principle recognizes that trust cannot be static in a dynamic digital environment. Access decisions should continuously adapt based on risk, taking into account contextual signals such as user and workload behavior, device health, location, vulnerability status and threat intelligence.

Rather than relying on fixed policies, organizations should ensure that trust levels reflect real-time conditions. This is particularly important as AI agents become more autonomous and gain access to enterprise systems and data. Security teams should continuously monitor AI agent activities and adjust permissions, tool access and levels of autonomy based on behavioral patterns, data sensitivity and emerging threats.

An AI agent that is considered low risk today may require additional controls tomorrow if its behavior changes or new risk indicators emerge. The goal is not simply to grant access securely, but to continuously validate that access remains appropriate as conditions evolve.

Gartner analysts will discuss key topics for CIOs and IT leaders, including AI, executive leadership, cybersecurity, operating models, and more, at the Gartner IT Symposium/Xpo, taking place in Kochi from 16-18 November 2026.

The author is Wayne Hankins, Sr Director Analyst at Gartner.

Disclaimer: The views expressed are solely of the author and ETCISO does not necessarily subscribe to it. ETCISO shall not be responsible for any damage caused to any person/organization directly or indirectly.

Join the community of 2M+ industry professionals.

Subscribe to Newsletter to get latest insights & analysis in your inbox.

All about ETCISO industry right on your smartphone!