In May 2026, Charter Communications (one of the largest telecommunications and broadband providers in the United States) disclosed a major data breach in which attackers allegedly used voice phishing (vishing) techniques to compromise an employee’s Microsoft Entra account and exfiltrate approximately 40 million customer records. This incident, among several others, continues to point to poorly managed identity management as a major vector for compromise. This notion was asserted in the CrowdStrike Global Threat Report 2026, which found that 82% of detections were malware-free, with adversaries relying on valid credentials, trusted identity flows and legitimate SaaS integrations to move through environments while evading traditional security controls.
Further compounding the pain for often stretched cybersecurity teams, the emergence of agentic AI has led to the exponential growth of Non-human Identities (NHI). In modern enterprise environments, these NHIs, encompassing service accounts, bots, API keys, OAuth tokens, and secrets, are estimated to outnumber human identities by a factor of 82:1 or more. Yet, despite the increasing number, data shows that only 52% of surveyed organizations have fully deployed Zero Trust architecture.
Zero Trust emerged in response to the erosion of traditional perimeter controls as organizations moved to cloud environments, SaaS platforms and increasingly distributed workforces. Its core principle was simple: never trust, always verify. Rather than relying on broad network boundaries, Zero Trust introduced stronger identity verification, least-privilege access, and micro-segmentation to limit the blast radius of a compromise. However, as AI initiatives increasingly require access across multiple systems, applications, and data sources, many traditional controls are coming under pressure. To remain effective in the era of agentic AI, organizations must evolve toward an identity-first approach to Zero Trust.
A perimeter control is the first line of defense for an asset. In every access scenario – be it a user on a workstation, an application server, or a SaaS API call – there is a source entity driven by an identity. By securing the identity through robust validation of the requesting party, cybersecurity teams can substantially reduce cyber risk exposure. In the section below, I discuss three highly effective controls to mitigate identity-related cyber threats, based on my 20-plus years of experience in transforming security in complex enterprises.
1. Transition to Behavior-Based Identity Authentication
The first step is to replace traditional multi-factor authentication (MFA) with behavior-based identity authentication. Once considered the gold standard control, traditional MFA can now be easily bypassed by sophisticated session hijacking, MFA fatigue attacks, SIM swapping, AI-driven social engineering and real-time phishing attacks that silently relay authentication codes.
By utilizing behavioral signals from unusual locations, time of day, biometrics, analyzing typing speed, mouse movements and login patterns, adaptive systems provide a means of continuous authentication. For instance, if a session’s risk score spikes due to anomalous behaviour, the system can dynamically terminate access or trigger a just-in-time challenge, transforming identity authentication from a gate into a continuous monitoring mechanism.
Here are three ways to implement behavior-based authentication in your organization:
- Deploy behavioral controls. Enable signals and integrate tools that continuously analyze usage patterns, typing patterns, mouse movements and login behavior to build a dynamic baseline for every user and detect anomalies in real time.
- Implement Identity Threat Detection and Response (ITDR). Ensure your team has the tooling and playbooks to detect and respond to identity-based attacks beyond the point of login, not just at the perimeter.
- Adopt Risk-Adaptive Access Control (RAAC). Configure access policies that automatically tighten or expand privileges based on evolving user behavior and contextual risk signals, rather than relying on static role assignments. For example, a hospital system may grant a surgeon full access to patient records during normal working hours but automatically restrict and challenge that same session if it suddenly attempts to export bulk records from an unrecognized location outside business hours.
2. Governing the ‘Agent Economy’
While agentic AI is radically streamlining processes, expanding revenue lines and cutting costs, it’s also become a major pain point for technology teams. These AI agents are no longer just tools; they are “digital employees” capable of executing contracts, committing spend and accessing sensitive CRM or financial data. This autonomy introduces new attack vectors. An AI agent with access to a financial system is an attack vector that can be hijacked through prompt injection, manipulated via poisoned data, or exploited through overprivileged access. To secure this new workforce, organizations must address the governance gap and adopt a layered, identity-first design:
- Enforce least-privilege access: apply role-based and attribute-based access controls to ensure agents can only reason and act within tightly defined boundaries, never beyond what the task needs.
- Issue and rotate short-lived credentials: treat every AI agent like a human identity, provisioning time-bound credentials for each session and revoking them immediately upon task completion.
- Maintain an immutable audit trail: log every action, API call and decision an agent executes so that anomalous behavior can be detected, investigated and attributed with precision.
3. Measure Zero Trust Maturity in the AI era
Zero Trust maturity in the AI era requires moving beyond static authentication. Cybersecurity teams must prioritise Just-in-Time (JIT) access and least-privilege principles. Furthermore, they must cryptographically verify every command an agent attempts to execute. A mature architecture employs Policy Decision Points (PDP) and Policy Enforcement Points (PEP) to limit the data and resource boundaries available to autonomous entities.
By implementing JIT access, organizations ensure that permissions are elevated only when needed and revoked immediately upon task completion. This “right-time, right-level” access model is the only way to effectively manage risk in a multi-cloud, multi-application world.
Here are three ways to mature your Zero Trust architecture for the AI era:
- Implement Just-in-Time access. Elevate permissions only when a task demands it and revoke them immediately upon completion, ensuring no agent or user carries standing privileges in perpetuity.
- Deploy Policy Decision and Enforcement Points (PDPs and PEPs) to evaluate every access request in real time, ensuring human and AI identities can access only the data, systems and resources required to perform their intended function.
- Cryptographically verify agent commands. Treat every instruction an AI agent attempts to execute as untrusted until verified, ensuring commands cannot be tampered with, replayed or injected by a malicious actor. For example, a hospital AI agent authorized to update medication dosages must have every command cryptographically signed and validated before execution, preventing an attacker from injecting a falsified instruction that silently alters a patient’s treatment.
Identity as the New Perimeter
The transition from hardware firewalls to a unified Identity Fabric is the cornerstone of modern digital trust. IAM is no longer a routine IT function. It is the foundational layer of enterprise security. In a world of cloud workloads and autonomous agents, identity is the only consistent control point.
Every agent, bot and service account must be treated as a privileged identity. IAM must be automated, continuous and intelligent.
As you finalize your technology roadmap, ask yourself: how are you governing the identities of the AI agents transacting on your behalf? And do you truly know which non-human accounts still hold the keys to your most sensitive data?
