Why flat networks hurt
The chapter opens with a network diagram showing a single subnet holding office laptops, finance servers, warehouse tablets, badge readers, an unpatchable printer and the loyalty database. In a flat network like this, an attacker who lands on one machine can reach every other target directly, so lateral movement costs almost nothing. Defenders are blinded too: when every connection is allowed, nothing looks unusual.
Segmentation divides the network into zones, typically user devices, servers, infrastructure and management, and untrusted or special-purpose devices such as guest Wi-Fi, IoT and production equipment. The author separates VLANs from segmentation: a VLAN only labels traffic, and control comes from firewall rules at the boundaries with default deny, so only explicitly allowed traffic crosses.
Deny rules can also log, turning blocked crossings into high-quality signals. To avoid self-inflicted outages, the chapter recommends defining zones and rules in log-only mode for a week or two, finding the unexpected legitimate flows, and only then enforcing. Host-based firewalls add a second boundary that travels with the device.
From castle-and-moat to zero trust
The castle-and-moat model trusts anything inside the perimeter. That no longer fits a world of remote work, SaaS and cloud data, unmanaged devices and attackers who are often already inside. Zero trust does not mean trusting nothing; it means granting trust per request, based on evidence. Its three principles are verify explicitly using identity, device health and context, use least privilege, and assume breach.
Zero trust is described as a direction to walk rather than a product to buy, assembled from identity and MFA, compliant endpoints, segmentation and logging. A conditional access policy that requires MFA, a compliant device and low sign-in risk for an internal application, wherever the user connects from, is an example. In the book's story, a contractor laptop that had fallen behind on patches was denied access to the ordering portal while sitting on the office network, which would have been allowed under the old model.
Micro-segmentation and east-west traffic
North-south traffic crosses the perimeter and has long been controlled by edge firewalls. East-west traffic flows between internal systems, has rarely been watched, and is exactly what attackers use to move. Micro-segmentation controls it at the level of individual workloads.
Start with the crown jewels. The example limits the loyalty database so that only two application servers can reach it on the database port, and logs every other attempt. Because nothing else should ever try, a hit on that deny rule is a near noise-free intrusion signal. The chapter warns against writing rules for every flow in the estate: micro-segment the few critical systems, such as finance, domain controllers and backup infrastructure, and leave broader zones coarse, because rules nobody maintains stop protecting anything.
DNS as risk and as control
Almost every connection starts with a DNS lookup. Malware uses DNS to find its command servers and sometimes to smuggle data out. A protective DNS resolver that refuses to resolve known malicious, phishing or newly registered domains blocks many threats before any payload arrives, and covers devices that cannot run an agent.
Two conditions make it work: clients must not be able to bypass the resolver with their own external or encrypted DNS, so outbound DNS is blocked except from your resolvers, and every query should be logged. Logs reveal tunnelling patterns and let you check later whether any machine contacted a domain later found to be malicious. DNS filtering does not help against direct IP connections or brand-new domains, so it is one layer among many.
Securing remote access
Every remote-access path should require phishing-resistant MFA where possible, a compliant device, least privilege and logging of each session. The author treats one rule as non-negotiable: never expose management interfaces such as remote desktop, SSH, database ports or hypervisor consoles directly to the internet; put them behind the access broker. Every remote-access grant should have an owner and an expiry date.
The expert section contrasts VPNs, which place a remote device on the network, with zero trust network access, which brokers access to individual applications per session after checking identity and device posture. ZTNA suits application access well but may not fit legacy non-web workloads, and it depends heavily on the strength of the identity provider and device-compliance signals.
KEY POINTS
Key points
- Draw your actual network and mark the crown jewels; if everything reaches everything, plan segmentation.
- Separate user devices, servers and management infrastructure with default-deny rules, starting in log-only mode.
- Micro-segment your most valuable system and make its deny rule log so crossings become alerts.
- Deploy a protective DNS resolver, block clients from bypassing it and log every query.
- Require MFA and a device-compliance check on every remote-access path with no trusted-LAN exception.
- Close any management port exposed to the internet and give every remote-access grant an owner and expiry.
QUESTIONS
Common questions
- Why is a flat network a security risk?
- In a flat network every device can reach every other device, so an attacker who compromises one laptop can move directly to servers and databases with no internal barrier. It also hides the attack, because all connections are permitted and none look unusual. Segmentation with default-deny rules between zones slows movement and makes blocked crossings visible in logs.
- What is the difference between a VLAN and network segmentation?
- A VLAN logically separates devices into different broadcast domains and address ranges, but it does not by itself control traffic between them. Segmentation that stops lateral movement requires firewall rules at the boundaries between VLANs, with default deny so only explicitly allowed traffic crosses. VLANs without enforcement are organised, not secure.
- Does zero trust require buying a new product?
- No. The Hardened Stack describes zero trust as an architecture and posture built from controls most organisations already have: strong identity and MFA, compliant devices, segmentation and logging. A practical first step is to put one internal application behind a policy requiring MFA and a compliant device, whether the user is in the office or remote.
- How should I start micro-segmentation with a small team?
- Start with the single most valuable system, such as a customer database, and allow only its legitimate callers on the ports they need, denying and logging everything else. Extend tight rules only to a few critical systems such as finance, domain controllers and backups. Writing rules for every flow creates a rule set a small team cannot keep current.
- What is protective DNS filtering and how do I make it effective?
- Protective DNS routes name lookups through a resolver that refuses to resolve known malicious or policy-violating domains, stopping connections before any payload is fetched. To make it effective, block outbound DNS except from your own resolvers so clients cannot bypass it, and log all queries to detect tunnelling and support later investigations.
- Is it safe to expose RDP or SSH to the internet temporarily?
- No. The Hardened Stack treats it as a non-negotiable rule that management interfaces such as remote desktop, SSH, database ports and hypervisor consoles are never exposed directly to the internet. Automated scans find them within minutes. Place them behind a remote-access broker that checks identity and device health first.