zero trust (ZT)

in English: zero trust security

advancedwidespread

A security principle stating that we trust no one and nothing in advance: every access must be verified separately each time.

In plain terms

Old corporate security worked like a castle: thick walls, one gate, and once someone got in, they could roam freely inside. Zero trust breaks with this. It is more like an office building where you must swipe your access card at every door, and only the doors leading to places where you actually have business open.

The basic idea is: 'never trust, always verify'. It does not matter whether you log in from the corporate network or from home, or whether you are a boss or an intern. The system checks with every request who you are, what device you are using, and whether you are entitled to what you are requesting.

It became important because the 'wall' disappeared: people work from home, data is in the cloud, and attackers often arrive 'from the inside' with stolen passwords. In the age of AI agents, it is even more important because the agent is also a participant in the system: it independently opens files and initiates operations, so it must be identified and restricted just like a human.

How it works

Identification: every user, device, program, and agent has its own verifiable identity. For humans, this typically means multi-step login; for programs and agents, it involves certificates or short-lived access keys.
Situation assessment: the system checks not only the password but also the circumstances, such as whether the device is known, the software is up to date, or the location and time are unusual.
Least privilege: everyone accesses only what is needed for the specific task and only for as long as necessary.
Isolation: the network and data are split into small, separately lockable parts. If an intruder breaches one point, they cannot easily proceed further.
Continuous monitoring: trust is not a one-time event. The system logs and re-verifies during operation, and revokes access in case of suspicious behavior.

An everyday example

In the morning, you open email from your corporate laptop and get in smoothly. In the afternoon, you try to open the payroll from your own phone on a café Wi-Fi: the system asks for further confirmation or simply denies access. You are the same person, but the situation is different, so the decision is different.

Who is affected

employeeThis is why you must confirm login more often, and why you see only company folders relevant to your work, not all of them. → Work & HR

entrepreneur, executiveAlongside remote work, cloud computing, and AI agents, this principle determines how much damage a single stolen password or hijacked agent can cause.

developerServices and agents you build need their own identifier and restricted permissions, not a single master key that opens everything. → Code

bank customer, investorYour bank partly for this reason requests additional confirmation for unusual transfers or logins from a new device. → Money

A common misconception

MythZero trust is a purchasable product that you install, and from that point on you are protected.

In realityThis is a mindset and design principle, not a single box or program. It is realized by multiple tools, rules, and organizational habits together, typically over years, step by step.

Limits and risks

Its implementation is expensive and slow, especially where many legacy systems exist that were not designed for such monitoring.
Excessive confirmation annoys people, who then look for loopholes, thereby weakening security.
It does not protect against everything: if an authorized user is deceived, or an agent is diverted by a hidden instruction, damage can occur just as easily within the granted permissions.
Continuous logging collects vast amounts of data on employee behavior, raising data protection and trust issues.

Similar, but not the same

VPN – A VPN grants access to the internal network through a protected tunnel, typically providing broad access there. Zero trust, by contrast, decides per resource and per request.
multi-step authentication – Multi-factor authentication is only one tool for verifying the identity of the entering person, while zero trust is the logic of the entire access system.
firewall – Traditionally, the firewall guards the network boundary. Zero trust starts from the assumption that no one can be blindly trusted even inside the boundary. The firewall can still be part of the implementation.

Related concepts

Broader

cybersecurity, information security, AI safety

Narrower

principle of least privilege, microsegmentation, multi-step authentication, agent identity

On the edge

access control, prompt injection, sandbox, safeguards, human approval

A brief history

The term itself first appeared in 1994 in Stephen Paul Marsh's doctoral dissertation on the computational modelling of trust. In the 2000s, a professional group called the Jericho Forum argued that the network "perimeter" offered less and less protection. The concept was widely popularised by John Kindervag, an analyst at Forrester, in 2010, with the slogan "never trust, always verify". Google began introducing its BeyondCorp solution around 2011 and, from 2014, described in several papers how it works at scale. The US National Institute of Standards and Technology (NIST) published the still-fundamental description of zero trust architecture (SP 800-207) in August 2020.

Where things stand

As of late 2026, this is the accepted security principle in large companies and public administration, although full implementation is still under way in most places. The latest question is how to extend the same rigor to autonomously working AI agents.

Status: 2026 November

In the news

Appears in 1 Agent Age stories; the curve shows weekly occurrence in the news magazine.

Security vendors build defences around coding agents: AWS Continuum, Wiz, SailPoint, Menlo, Snyk, MicrosoftAug 3, 2026 · Code

Also known as

zero trust · zero trust · zero-trust model · nulla bizalom · zero trust security · zero trust architecture · ZTA

Sources

This entry was machine-edited and has undergone machine expert review; the “In the news” section updates automatically from Agent Age stories.