Redefining Zero Trust in the age of Agentic AI

Agentic AI makes Zero Trust practical at cloud scale by autonomously verifying identities and devices, micro-segmenting access and adapting trust in real time
 
7 minutes 所要時間
Nicholas Ismail
Nicholas Ismail
Global Head of Brand Journalism, HCLTech
7 minutes 所要時間
共有
記事を聴く
ミュート
30秒戻る
30秒進む
Redefining Zero Trust in the age of Agentic AI

is a cybersecurity framework that requires continuous verification of users, devices, and systems before granting access to business resources. But in fast-changing cloud environments, manual policy updates and approvals simply can’t keep up.

, or self-driving software agents with bounded autonomy, can automate identity verification, micro-segmentation and adaptive trust scoring to reduce lateral movement and ensure only verified users and devices touch critical assets.

  • Zero Trust: Never trust, always verify, enforced continuously
  • Agentic AI: Autonomous agents that decide and act (with guardrails), not just assist
  • The problem: Manual enforcement lags behind cloud scale and speed
  • The solution: Agentic AI automates and synchronizes Zero Trust controls in real time

What is Zero Trust and why does it matter?

“Zero Trust isn’t something you buy; it’s an architectural mindset you apply, often differently across environments and levels of regulation” - Prikshit Goel, Global Practice Head, Cybersecurity, HCLTech

Defining Zero Trust Principles

Zero Trust (ZT) is a security model that assumes no implicit trust, inside or outside the network. Every request is authenticated, authorized and encrypted based on context.

Core principles:

  • Verify explicitly: Authenticate and authorize every user, device and workload continuously
  • Least privilege access: Grant the minimum access needed, tighten dynamically
  • Assume breach: Contain damage with segmentation and continual monitoring
  • Context-aware decisions: Use identity, device health, location, behavior and risk signals

Key challenges in traditional Zero Trust implementation

1) Manual policy drag: Writing, tuning and rolling out policies is time-consuming and error-prone, especially across multiple clouds and business units. Small misconfigurations can create large exposure windows.

2) Dynamic cloud complexity: Workloads spin up and down, identities proliferate and APIs change daily. Keeping micro-segments, routes and access rules in sync manually is unrealistic

3) Signal overload: Identity, EDR, network and SaaS logs generate more data than human teams can triage quickly, delaying detection and response

A simple example: after a merger, two cloud estates must be segmented while enabling collaboration. Manual firewall and IAM updates can take weeks. Meanwhile, attackers need minutes to laterally move if they gain a foothold.

What is Agentic AI and how does it enhance Zero Trust?

What is Agentic AI?

Agentic AI are autonomous, goal-driven software agents that perceive signals, make decisions within defined guardrails and execute tasks. They should escalate decisions to humans when confidence is low or impact is high.

Key characteristics:

  • Autonomy with bounds: Acts within pre-approved playbooks and risk limits
  • Continuous learning: Improves policies via feedback and outcomes
  • Tool-use and orchestration: Calls APIs (IAM, EDR, SD-WAN, CSPM) to implement changes
  • Human-on-the-loop: Operators supervise, approve sensitive actions and set guardrails

How Agentic AI automates Zero Trust processes

  1. Identity verification: AI agents correlate login context, such as user, device posture, location and behavior, with recent activity to step up authentication, quarantine a device or block a session, automatically.
  2. Micro-segmentation: Agents observe traffic patterns, recommend segment boundaries, create/adjust policies and verify reachability, which shrinks the blast radius without endless manual rule-writing.
  3. Adaptive trust scoring. Agents compute per-entity risk scores, including user, device and workload, in real time and adjust privileges; tightening access at the first sign of drift or compromise.

How does Agentic AI address security challenges?

“Agentic AI synchronizes outcomes across multicloud controls by addressing the complexity that hinders manual methods. Real agents act within guardrails, and domain-specific models bound non-determinism, so humans stay on the loop. In regulated environments, this is the viable path to autonomy, with fewer false positives and no false negatives” - Prikshit Goel, Global Practice Head, Cybersecurity, HCLTech

Blocking lateral movement with micro-segmentation

Micro-segmentation isolates applications and data so that a compromised account or host can’t traverse the environment. Agentic AI accelerates this by:

  • Mapping dependencies automatically, such as who talks to what, how often and why
  • Proposing and enforcing least-privilege flows
  • Watching for policy drift and instantly rolling back if outcomes deviate

As an example: An anomalous service-to-database call appears in a non-production subnet. The agent flags the flow, lowers the service’s trust score, creates a temporary deny rule for that path and notifies an operator, which stops potential lateral movement in seconds.

Ensuring continuous verification of users and devices

Traditional verification is periodic, while AI-driven verification is continuous.

Capability

Manual (Traditional)

AI-Driven (Agentic)

Risk evaluation Static rules at login Real-time, multi-signal scoring
Response speed Minutes–days (tickets) Seconds (automated actions)
Policy updates Batch, brittle Continuous, outcome-based
Consistency Varies by team Standardized across clouds

Adaptive trust scoring for dynamic cloud environments

Adaptive trust uses live context to dial access up or down.

Signals commonly factored in:

  • Identity assurance (MFA strength, credentials age)
  • Device posture (patch level, EDR health, jailbreak/root status)
  • Behavior analytics (time, location, velocity, anomalies)
  • Data sensitivity (PII, financials, IP)
  • Workload health (CVEs, misconfigurations, drift)
  • Threat intel (active campaigns, IOCs)

When risk rises, agents automatically and reversibly trigger step-up authentication, session restriction, micro-isolation or revocation.

What are the benefits of combining Zero Trust and Agentic AI? 

1. Enhanced Security Posture

  • Faster containment: Automated segmentation and revocation limit the impact of a breach
  • Fewer blind spots: Agents correlate across identity, device, network and workload
  • Outcome consistency. Policies are enforced uniformly across multicloud and on-prem

"Zero Trust is an architectural mindset, not a single product. Agentic AI shines by synchronizing outcomes across different controls and cloud providers, keeping decisions consistent as environments change” - Prikshit Goel, Global Practice Head, Cybersecurity, HCLTech 

2. Operational efficiency

  • Lower manual toil: Tickets, rule updates and routine verifications are automated
  • Faster policy rollouts: Agents propose, simulate and deploy with safe rollback
  • Reduced alert fatigue: Fewer false positives reach analysts; true positives get richer context

Scalability for complex environments

“Agentic AI handles thousands of identities, devices and workloads by learning normal patterns and enforcing guardrails at scale, which is ideal for multicloud, multi-region operations and regulated subsidiaries” - Prikshit Goel, Global Practice Head, Cybersecurity, HCLTech

What are the challenges of using Agentic AI for Zero Trust?

Agentic AI security risks

  • Non-determinism: Generative models can vary in decisions. Without bounds, outcomes may drift
  • Misconfiguration risk: Poor guardrails or training data can lead to over-blocking or gaps
  • Model exposure: Data leakage or prompt injection against operational agents
  • Ethical concerns: Autonomy without transparency or recourse erodes trust

“To overcome these challenges, constrain agents with domain-specific models (often “small” language models) and curated security data. This narrows decision variance to a known, reasonable range, enabling safe autonomy with human oversight” - Prikshit Goel, Global Practice Head, Cybersecurity, HCLTech

 

Future of Payments Report 2025

 

AI governance in Zero Trust frameworks

Why it matters: Regulators and boards expect accountability for autonomous decisions affecting security and privacy.

Governance essentials:

  • Purpose and scope: Define what the agent is allowed to decide and do
  • Model strategy: Prefer domain-specific models, document training data and synthetic data use
  • Guardrails: Confidence thresholds, change windows, kill switches and automatic rollback
  • Human oversight: Human-on-the-loop for sensitive actions and auditable approvals
  • Risk and testing: Pre-deployment simulation, red-teaming and ongoing drift monitoring
  • Transparency and logs: Tamper-evident records of inputs, decisions and outcomes
  • Compliance alignment: Map to frameworks and AI management practices

Practical steps for implementing Zero Trust reinforced by Agentic AI

Key steps for organizations

  1. Run a Zero Trust readiness assessment 
    Assess inventory identities, devices, apps, data and flows. Identify crown-jewel assets and current controls.
  2. Target high-impact automation candidates
    Pick use cases with clear wins: adaptive MFA, session containment or automated micro-segmentation for a critical app.
  3. Deploy Agentic AI for identity and segmentation 
    Start in monitor-only mode, simulate changes and then enable enforcement with rollback. Integrate with IAM, EDR, network and cloud APIs.
  4. Operationalize adaptive trust scoring 
     Define risk factors and thresholds and calibrate to reduce false positives while avoiding false negatives.
  5. Stand up AI governance 
    Establish guardrails, audit trails, approval flows and model lifecycle management, which are aligned to the organization’s privacy posture and regulatory obligations.

KPIs to watch: Mean time to contain (MTTC), policy drift rate, false positive/negative rates, percentage of automated decisions reviewed and blast radius reduction for the most sensitive systems.

Key takeaways

  • Zero Trust is an architectural mindset, not a product. Manual enforcement struggles at cloud scale
  • Agentic AI automates the hard parts, including identity checks, micro-segmentation and adaptive trust
  • Bounded autonomy is essential: Use domain-specific models, guardrails and human-on-the-loop oversight
  • Governance makes it safe: Clear scope, logging, testing and rollback build confidence for regulated environments
  • Start small, scale fast: Pilot high-impact use cases, measure outcomes and then expand across clouds and business units.

FAQs

What is Zero Trust in cybersecurity?
A security framework that never assumes trust. Every access request is verified continuously based on identity, device, behavior and risk.

How does Agentic AI improve Zero Trust security?
By automating identity verification, micro-segmentation and adaptive trust scoring, and adjusting policies in real time to limit lateral movement.

What are the benefits of combining Zero Trust with Agentic AI?
Stronger protection, fewer false positives, faster containment and lower operational toil at cloud scale.

Are there risks associated with using Agentic AI for Zero Trust?
Yes. Non-deterministic decisions, misconfigurations and ethical concerns require guardrails, governance and oversight.

How can organizations implement Agentic AI in a Zero Trust framework?
Assess readiness, select targeted automations, deploy agents with simulation and rollback, tune trust scoring and establish governance.

What is micro-segmentation in Zero Trust security?
The practice of isolating workloads and flows so an intrusion can’t spread: shrinking the blast radius.

Why is AI governance important in Zero Trust frameworks?
Governance provides accountability, compliance and safety, ensuring autonomous actions are transparent, auditable and reversible.

共有:
_ Cancel

お問い合わせ

Want more information? Let’s connect