One forgotten maintenance account. No multi-factor authentication. No access controls scoped to what that account actually needed. That was the entire attack surface that allowed threat actors to breach PowerSchool in early 2025 - exposing the personal data of 62 million students and nearly 10 million teachers across North America in what became one of the largest education-sector breaches ever recorded.
No exploit chain. No zero-day. Just one credential used against a system that trusted it completely because that is what the architecture was designed to do.
This is the threat pattern that dominated 2025 and is defining 2026. Sophos analyzed 661 incident response cases across 70 countries and found that 67% of all cybersecurity incidents were rooted in identity-related attacks - compromised credentials, brute-force access, phishing, and stolen authentication tokens. Unit 42's Global Incident Response Report 2026 found that in 87% of cases, attacker activity crossed multiple attack surfaces simultaneously - endpoints, identity systems, networks, and cloud services within a single intrusion. And in almost 90% of investigations, identity weaknesses played a material role.
Credential abuse has now overtaken malware as the primary intrusion method. Brute-force attacks are drawing level with exploitation as an initial access technique. Dwell time - the period between initial compromise and detection - has dropped to a median of three days. Not because defenders are improving. Because attackers are moving faster.
The architectural response to this threat environment has a name: Zero Trust Network Access - ZTNA.
What ZTNA Actually Means
ZTNA is a security framework that replaces location-based trust with identity and context-based access control. Its foundational principle is absolute: no user, device, or network connection is trusted by default - and trust, once earned, is not permanent.
Where traditional access models authenticate a user at the perimeter and then grant them broad access to network resources, ZTNA treats every access request as unverified regardless of where it originates.
A user inside the corporate building makes a request. A contractor connecting from home makes a request. An API running an automated workflow makes a request. All three are evaluated against the same criteria: identity, device health, behavioral context, and the specific resource being requested.
Access is granted to that specific application - and nothing else. The broader network is invisible. Adjacent systems are unreachable. The blast radius of any compromised identity is contained, by design, to the applications that identity was authorized to access.
NIST SP 800-207 formalized Zero Trust architecture as the recognized standard for enterprise security. Gartner defines ZTNA as products and services that "create an identity and context-based, logical-access boundary that encompasses an enterprise user and an internally hosted application or set of applications. The applications are hidden from discovery, and access is restricted via a trust broker to a collection of named entities, which limits lateral movement within a network."
In simpler terms: attackers who gain a credential cannot use it to roam. They reach what that credential is authorized to reach - and that is the end of the breach's expansion.
The Problem ZTNA Solves That VPNs Cannot
The PowerSchool incident is instructive not because it was sophisticated, but because it was not. No zero-day exploit. No advanced persistent threat toolkit. One forgotten account with valid credentials and no MFA - used against an architecture that was designed to trust it.
VPNs operate on exactly this model. Authentication happens once. Network access follows automatically. The breadth of that access depends on how carefully administrators have configured network segmentation - which, in practice across most enterprise environments, is not carefully enough.
A recent CSO finding underscores how widespread this exposure is: 91% of enterprise users log in at their highest level of privilege and stay there. Not because of negligence, but because most access architectures do not dynamically adjust privilege based on what a user actually needs at any given moment. They grant maximum standing access at login and leave it in place indefinitely.
This standing privilege - always-on, rarely audited, seldom adjusted - is the structural condition that makes identity compromise so valuable to attackers. One credential in a flat, over-privileged environment is not a foothold. It is a master key.
ZTNA removes standing privilege from the equation. Access is not granted at login and maintained for the duration of the session. It is evaluated continuously against the current state of the user, the device, and the behavioral context - and adjusted dynamically as any of those variables change.
The Key ZTNA Features That Define Effective Implementations
Understanding what ZTNA features actually do in production separates platforms that eliminate VPN-era risk from those that replicate it with a different name.
1. Contextual Access Control Beyond Identity
Identity verification is necessary. It is not sufficient. The PowerSchool attacker had a valid identity. What they lacked was a system that evaluated context alongside identity - device posture, behavioral baseline, access pattern consistency, and the specific resource being requested.
Effective ZTNA platforms evaluate six contextual signals simultaneously: user identity, group membership and role, device security posture, geographic location, time of access relative to established patterns, and the sensitivity of the resource being requested. Any anomaly in any signal is a reason to restrict or deny access - not to flag it for review after the fact.
2. Non-Human Identity Governance
This is the ZTNA feature that most platforms have not yet adequately addressed - and the one that 2026 threat intelligence has elevated to critical status. The 2026 Identity Exposure Report identified 18.1 million exposed API keys in enterprise environments. APIs, service accounts, automation agents, and increasingly, Agentic AI systems are all operating with enterprise credentials, making access requests, and modifying systems - often with permissions that were provisioned once and never reviewed.
An API connecting with valid credentials from a compromised host should not carry the same trust as one operating from a clean, policy-compliant environment. Non-human identities require the same continuous evaluation as human ones: identity verified, host posture assessed, action authorized, and the entire interaction logged for audit.
3. Dynamic Policy Adjustment - Not Static Rules
Most enterprise access policies are written once and revisited annually, if at all. In environments where user roles change, devices drift in and out of compliance, and threat intelligence changes daily, static policies are not security - they are documentation.
ZTNA platforms that enforce dynamic policy adjustment update access rights in real time based on changes in any contextual variable. A device that was compliant at 9 AM and has its endpoint protection disabled by 11 AM is not the same device. The session that granted access at 9 AM should not still be providing the same access at 11 AM. Privileges shrink as risk rises - automatically, without an IT ticket and without interrupting the session of every other user connected at that moment.
4. Visibility-First Architecture for Internal Environments
Most ZTNA discussions focus on remote access. The blind spot is the internal network. Lateral movement - the technique used in the vast majority of breaches that turn an initial compromise into a systemic incident - happens inside the network, between systems that already trust each other because they are on the same segment.
ZTNA applied to internal environments enforces identity-aware segmentation at the LAN edge - governing traffic not just from outside the perimeter inward, but between devices within the same campus or branch network. Zero Trust does not stop at the corporate gateway. It applies to every connection, including the ones from devices that are already inside.
5. Application Invisibility
One of the most underappreciated ZTNA features is what it does before any access is attempted: it makes applications invisible to unauthorized users. In a ZTNA architecture, applications are not internet-facing services with open ports waiting to be probed. They exist behind a trust broker that requires verified identity and compliant device posture before any connection is established. Attackers cannot scan for what they cannot discover.
This "dark cloud" architecture eliminates an entire class of reconnaissance that precedes most attacks - the mapping of which services exist, which ports are open, and which endpoints are reachable. Remove the ability to discover, and you remove the ability to target.
6. Agentless Access for Unmanaged and Third-Party Endpoints
Identity-based access controls must extend to the devices that cannot run a security agent: contractor laptops, partner devices, BYOD endpoints, and increasingly, IoT and OT systems. Agentless ZTNA enforces access policies through browser-based sessions or lightweight connectors - scoping access to the authorized application without requiring visibility into or control over the device's full configuration. Third parties get precisely what their role requires. Nothing more is reachable. Nothing more needs to be secured on their behalf.
Also Read: Zero Trust Network Access (ZTNA): Secure, Controlled, and Deployed from Everywhere
What These Features Prevent: Three Recent Incidents
The value of ZTNA features is most clearly visible when measured against what happens without them.
PowerSchool (Early 2025): A single unprotected maintenance credential without MFA exposed 62 million student records and nearly 10 million teacher records. In a ZTNA architecture, that maintenance account would have had access only to the specific maintenance interface it required - not the broader database environment its network location made reachable. Application invisibility would have prevented the attacker from discovering what else existed. Dynamic policy adjustment would have flagged the unusual access pattern. The scope of the breach would have been contained to the surface area of one account's authorized access.
Scattered Lapsus$ / Scattered Lapsus$ Hunters (2025): This group bypassed MFA entirely by exploiting OAuth authorization flows - convincing users to authorize legitimate-looking applications that granted the attackers long-lived session tokens. Standard ZTNA would not have stopped the OAuth manipulation. But ZTNA platforms with behavioral analytics would have flagged the resulting access patterns - unusual volumes of data accessed, applications reached outside the user's normal behavioral baseline, access from unexpected locations - and restricted the session before exfiltration reached its full scale.
Unit 42 2026 Incident Response Data: In 87% of cases investigated, attacker activity crossed multiple attack surfaces within a single intrusion - a pattern consistent with lateral movement through environments where internal trust was implicit. ZTNA's application-level micro-segmentation directly addresses this: even with valid credentials, a user can only reach the applications their role permits. Movement to adjacent systems requires re-verification. The multi-surface pattern that Unit 42 repeatedly documented is, by design, blocked before the second pivot.
Our Recommendation: Versa Networks
The threat intelligence of 2025 and 2026 has clarified what a serious ZTNA provider must deliver: contextual access control that covers non-human as well as human identities, continuous dynamic policy adjustment that adapts to session-level changes, internal visibility that enforces Zero Trust at the LAN edge and not just the perimeter, and an architecture that makes applications invisible to unauthorized discovery.
Versa Networks delivers all of this - and has extended it into a domain that no other major ZTNA platform has yet addressed at production scale: the governance of Agentic AI.
As enterprises deploy AI agents that interact directly with infrastructure - running queries, enforcing policies, modifying configurations through Model Context Protocol servers - the trust model extends to non-human actors operating with enterprise-level permissions. Versa launched the industry-first patented Zero Trust MCP Server to address precisely this gap. Every AI agent action flows through VersaONE's management plane, governed by role-based access controls, tenant authorization, and fully auditable logs. AI agents are not assumed to be trustworthy because they were deployed internally. They are evaluated against the same Zero Trust principles as every human user: identity verified, action authorized, interaction logged.
For internal network environments where lateral movement is the primary risk, Versa's VSPA (Versa Secure Private Access) delivers a visibility-first ZTNA model that prevents lateral movement, enforces identity-aware segmentation at the LAN edge, and ensures compliance across internal environments with single-pass enforcement. Zero Trust does not stop at the enterprise gateway in Versa's architecture. It applies from the campus LAN to the cloud workload - continuously.
The contextual Zero Trust model Versa implements treats device posture not as a login checkpoint but as a live variable. Endpoint Information Profiles update throughout every session. If malware is detected, a configuration drift is flagged, or a patch falls out of compliance during an active session, access policies adapt immediately - privileges shrink without disconnecting the session. For non-human identities, the same applies: an API or automation agent connecting from a host with degraded security posture receives reduced trust, regardless of whether its credentials are valid.
A concrete production outcome validates this architecture. T-Mobile Czech Republic used Versa technology to lead a Tier-1 financial institution through the transition from legacy VPNs to a modern, identity-centric SASE architecture - deploying in-country ZTNA Points of Presence with Zero Trust enforced across complex, multi-platform banking environments on an accelerated timeline. Financial institutions operate under some of the most stringent access control requirements in any regulated industry. The fact that this deployment was executed successfully at a major bank reflects the production maturity of Versa's ZTNA implementation.
Identity has become the primary attack surface of modern enterprise security. The evidence from Sophos, Unit 42, and the PowerSchool incident all point to the same architectural gap: access models that trust credentials at login and maintain that trust without continuous evaluation. ZTNA closes that gap - not by adding more authentication friction, but by making trust proportional, dynamic, and bounded by what the identity actually needs. Versa has built that model, extended it to non-human identities and Agentic AI, and validated it in production at the highest levels of security sensitivity.
Top comments (0)