Zero Trust Architecture

Intermediate · 4 min

Why "never trust, always verify" replaced the perimeter model, and what actually enforces it: identity, device posture, and micro-segmentation.

Before thisThe OWASP Top 10 for Architects

IBM Technology · 4 min

Zero Trust Architecture

The player loads from youtube-nocookie.com when you press play. Nothing reaches YouTube before that.

Taught by IBM TechnologyReport a problem with this video
Not started
Set this lesson's status

What you'll learn

  • The perimeter model assumes everything inside the corporate network is trustworthy and everything outside is not; zero trust assumes neither — every request is verified on its own merits, regardless of where it originates.
  • "Never trust, always verify" is enforced continuously, not once at login — a zero-trust system re-evaluates identity, device posture, and context (location, time, behavior) on every request, not just at the start of a session.
  • Micro-segmentation breaks a network into small, independently controlled zones so that compromising one service doesn't grant an attacker lateral access to everything else — the architectural opposite of one flat network where anything inside can reach anything else.
  • Identity federation (SAML/OAuth/OIDC, covered in system-design's authn-and-federated-identity-at-scale) is a building block zero trust relies on, not a replacement for it — proving who is asking is necessary but not sufficient without also checking device health and enforcing least-privilege access.
  • Zero trust is a model and a set of enforcement points (policy decision points, policy enforcement points), not a single product — a vendor selling "a zero trust solution" is selling one piece of an architecture an organization still has to assemble and operate.

The model the perimeter replaced

For decades, enterprise network security worked like a castle: a hardened perimeter (firewalls, VPN gateways) separated a trusted inside from an untrusted outside, and once a device or user was inside — on the corporate network, through the VPN — it was broadly trusted to reach most of what was there. This worked reasonably well when “inside” meant a physical office network. It stopped working as employees went remote, applications moved to clouds the company didn’t operate, and attackers got good at one specific move: compromise a single device or credential at the edge, then move laterally across the trusted interior, which had comparatively little internal friction to stop them.

Zero trust rejects the premise. Instead of “trusted inside, untrusted outside,” every request is verified on its own merits regardless of where it originates — a request from an employee’s laptop on the corporate Wi-Fi gets exactly as much scrutiny as one from a coffee shop. The shorthand for this is “never trust, always verify.”

Verification is continuous, not a one-time gate

The perimeter model’s trust decision happens once: authenticate at the VPN gateway or the office door, then move relatively freely. Zero trust’s trust decision is re-evaluated continuously — on every request, or close to it. A policy decision point considers several signals together before granting access to a specific resource:

  • Identity — who is asking, authenticated with strong multi-factor credentials, not just a password.
  • Device posture — is the requesting device managed, patched, running expected security software, not jailbroken or showing signs of compromise?
  • Context — is this request consistent with the user’s normal location, time of day, and behavior, or does it look anomalous?

Identity federation protocols — SAML, OAuth, OIDC, covered in system-design’s authn-and-federated-identity-at-scale — are the mechanism that answers the “who” question, and zero trust depends on them. But identity alone isn’t zero trust: a valid, correctly authenticated user on a compromised or non-compliant device is still a risk zero trust is designed to catch, which is why device posture and context sit alongside identity rather than being replaced by it.

Least privilege and micro-segmentation

Verifying identity and context on every request only helps if what’s granted afterward is narrow. Zero trust pairs continuous verification with least-privilege access: a verified request gets access to exactly the specific resource it asked for, for as long as it needs it, not standing access to a broad network segment. Micro-segmentation is how this gets enforced structurally — instead of one flat network where any internal host can reach any other, the network (or the service mesh, in a cloud-native system) is divided into small zones, each with its own enforced policy about what can talk to what. If an attacker compromises one service, micro-segmentation is what stops that compromise from becoming lateral movement into everything else; without it, the blast radius of a single breach is the whole network, which is precisely the perimeter model’s failure mode zero trust exists to close.

What zero trust is not

It’s tempting to read “zero trust” as a product category, because plenty of vendors sell one under that label. It’s more accurate to treat it as an architectural model implemented through a combination of pieces an organization assembles: identity providers, device-management and posture-checking tools, policy decision and enforcement points, and a segmented network or service mesh that actually carries out the per-request decisions those pieces make. Buying one vendor’s “zero trust” product typically covers one piece of this — strong identity, say, or network segmentation — not the whole model. It’s also distinct from network-layer defenses like DDoS mitigation (system-design’s ddos-mitigation-and-rate-limiting-for-security), which protect availability against volumetric attacks; zero trust is about authorization and access decisions for legitimate-looking traffic, a different layer of the same overall security posture.

Why this is an architectural decision, not a tooling purchase

Adopting zero trust means deciding, system by system, what the smallest useful segmentation boundary is, which signals a policy decision point actually needs to evaluate, and how much friction continuous verification adds to legitimate use versus how much lateral-movement risk it closes off. None of that is answered by a product datasheet — it’s a design exercise an architect does once per system, the same way choosing an architectural style or a data model is, and it’s exactly the kind of decision this roadmap’s architecture-decision-records-and-tradeoffs lesson is about recording well.

Check yourself

  1. 01What's the core assumption zero trust rejects that the perimeter model depends on?

    The perimeter model trusts a request based on where it comes from — inside the corporate network is safe, outside is not; zero trust rejects that assumption entirely and verifies every request on identity, device posture, and context regardless of network location.

  2. 02Why does zero trust verify continuously instead of just at login?

    A session that was legitimate at login can become compromised afterward — a stolen session token, a device that gets infected mid-session — so continuous verification re-checks identity and context on an ongoing basis rather than issuing one blanket trust decision that lasts for the whole session.

  3. 03How does micro-segmentation limit the damage of a single compromised service?

    Without segmentation, one compromised service on a flat network can reach every other service on that network; micro-segmentation puts each service or small group of services in its own enforced zone, so a compromise is contained to what that zone can reach, not the whole network.

Go deeper

Notes

Private to you

Esc
↑↓ navigate↵ open? shortcuts

Keyboard shortcuts

Press Esc to close

Anywhere

⌘K
Search and commands
/
Search
?
This sheet
Esc
Close

Reading a lesson

J
Next lesson
K
Previous lesson
B
Bookmark