Blog/Vulnerabilities

Network Security Fundamentals That Still Decide Incidents

Why flat networks turn one compromised laptop into a company-wide incident, and how segmentation, firewalls, IDS, DNS controls and monitoring help.

CyberWatch AI13 November 2026 · 8 min read
A network switch with many ethernet cables plugged into a wall-mounted rack

A marketing laptop opens an attachment it should not have. Within forty minutes the attacker has read the credentials cached on that laptop, used them to log into a file server, found a spreadsheet of service account passwords, logged into the virtualization management console, and started deleting backups.

Nothing in that sequence required skill. It required a network where any device could talk to any other device on any port, which describes most internal networks that grew organically. The attacker did not break through anything after the first step, because there was nothing left to break through.

Network security has been declared obsolete repeatedly as workloads moved to cloud and staff moved home. What actually happened is that its job narrowed and got more important: it is the control that decides how far an intrusion travels.

Segmentation: the one that matters most

Segmentation divides a network into zones and controls what crosses between them. CISA's guidance on layering network security through segmentation makes the case in a page, and the underlying logic is simple. If the finance server and the reception iPad are in the same unrestricted space, you have one security zone with the security of its weakest member.

A warning about a common misunderstanding: VLANs separate traffic logically, but they only provide security if routing between them is filtered. A network with fifteen VLANs that all route freely to each other is a well-labeled flat network. The boundary is the filtering, not the VLAN.

Zones that pay off first

You do not need per-workload microsegmentation to get most of the benefit. These divisions are achievable and each removes a category of incident.

  • Management interfaces in their own zone. Hypervisor consoles, switch and firewall administration, out-of-band management controllers, backup system consoles. These should be reachable only from a small set of administrative sources, never from an ordinary user device, and never from the internet. The spreadsheet-to-virtualization-console step in the opening example is impossible if this zone exists.
  • End-user devices separate from servers, with only the specific services users need crossing the boundary.
  • Devices that cannot be patched in isolation. Printers, cameras, building controls, medical and industrial equipment, anything running an operating system the vendor abandoned. These will never be secure, so the answer is to limit what they can reach.
  • Guest and personal devices nowhere near the corporate network, which is usually the easiest win available.
  • Backup infrastructure reachable only from what it needs to back up, and not administrable with the same credentials as everything else. Backup deletion is a standard ransomware step, and a backup an attacker can delete is not a backup.

Client isolation on wireless networks deserves a mention here too, since it stops devices on the same network from reaching each other directly and costs nothing to enable.

How to get there from a flat network

The approach that works is incremental and evidence-led. Observe actual traffic first, using flow logs or a span port, because nobody knows what the network is really doing and the guesses are always wrong. Pick one high-value system and segment that, learning the process on something manageable. Write rules that allow what you observed plus what the application owner can justify, then watch what breaks in a logging-only mode before enforcing. Expand one zone at a time.

The approach that fails is a comprehensive segmentation design produced up front, which collides with undocumented dependencies, and gets rescued by a rule allowing everything to reach everything "temporarily". Our zero trust guide covers the identity-centered version of the same goal, which is where this work leads.

Firewalls: two different jobs

NIST's SP 800-41 Rev. 1 remains a clear treatment of firewall policy, and its central point is that a firewall is only as good as the rule base, which decays.

At the edge, the firewall decides what the internet can reach. The rule to apply is default deny inbound, with a short, justified list of exceptions. Almost every mass-exploitation event lands on something that was exposed and did not need to be: a management portal, a file transfer appliance, a remote access gateway, a database port. The highest-value review you can run is to list every inbound rule and ask who owns it and why it still exists.

Outbound filtering is the half that gets skipped, and it is where you catch things. Most intrusions need to reach outside at some point, to retrieve tooling, receive commands or send data out. A network where any device can connect anywhere on any port gives that away for free. Restricting outbound traffic to what is needed, forcing web traffic through a proxy, and blocking direct outbound DNS except from your own resolvers all make an attacker's job noticeably harder.

Internally, firewalls enforce the segment boundaries described above. This is the most underused firewall role in most organizations.

Two maintenance habits matter more than the product. Review rules on a schedule and delete the ones nobody can explain, because rule bases accumulate and never shrink on their own. And keep the firewall's own management interface off the internet and its firmware current, since network edge devices have repeatedly been the entry point for serious intrusions.

Intrusion detection and prevention

NIST's SP 800-94 is the reference text here and sets expectations well. An intrusion detection system watches traffic and alerts; an intrusion prevention system sits in the path and can block. The engine is frequently the same, and the decision is about how much you trust it to interrupt real traffic.

What these systems are good at is catching known exploitation attempts against known vulnerabilities, scanning and reconnaissance, traffic to known malicious infrastructure, and protocol behavior that should not occur. What they are weaker at is encrypted traffic, which is now most traffic, and anything that looks like legitimate use of legitimate services.

That second limitation has pushed the useful work toward metadata and flow analysis: who connected to whom, how often, how much data moved, at what hours. You do not need to read the content of a connection to notice that a workstation has been talking to one external address every five minutes for a week, or that a server which never initiates outbound connections suddenly uploaded two gigabytes at 3am.

Inspecting encrypted traffic by terminating and re-encrypting it is possible and carries real costs: privacy implications, certificate management, performance, breakage of applications that pin certificates, and the creation of a high-value interception point. It can be justified for specific segments and is rarely worth it everywhere.

DNS security, the cheapest control available

Nearly every attack involves a name lookup. Phishing links resolve a domain. Malware resolves its command and control domain. Data exfiltration to a cloud service resolves a hostname. That makes DNS both a control point and an excellent source of evidence.

A protective DNS resolver refuses to answer queries for known malicious domains, so the connection fails before it starts. CISA operates such a service for its constituents and several commercial and community options exist. It is inexpensive, quick to deploy, and catches a real share of attacks at exactly the moment a user has already clicked.

Two configuration points make it work rather than decorate the network. Force all DNS through your chosen resolvers by blocking direct outbound DNS from everything else, otherwise a device simply uses another resolver and your filtering is advisory. And log the queries, because DNS logs are among the most useful records you can keep: they show which machine looked up a malicious domain and when, which is often the first solid fact in an investigation.

Protect your own DNS records too. Registrar account takeover lets an attacker redirect your domain wholesale, so registrar accounts deserve strong multi-factor authentication and registry lock where offered. Keeping an eye on lookalike domains registered against your brand belongs in the same routine.

Monitoring: what to actually collect

Insufficient internal network monitoring appears on the NSA and CISA list of the most common misconfigurations found in large networks, and it is usually not for lack of logs. It is that the logs exist in places nobody looks.

NIST's SP 800-92 covers log management properly, and the NCSC's material on building a security operations centre is realistic about what a small team can sustain. Rather than collecting everything, start with the sources that answer investigative questions.

  • DNS query logs, which tell you what a device tried to reach.
  • Network flow records, which show who talked to whom and how much moved.
  • Firewall deny logs at segment boundaries, which reveal lateral movement attempts.
  • Authentication logs, especially failures and logins from unusual sources.
  • Proxy or web gateway logs, which give you URLs rather than just addresses.
  • VPN and remote access logs, including session duration and source country.

Then alert on a short list of things that should rarely happen: a device scanning multiple internal hosts, authentication attempts against many accounts from one source, outbound connections from servers that have never made them, large uploads to file sharing services, connections to administrative interfaces from outside the administrative zone, and any traffic that a segment boundary should have made impossible. Our threat intelligence guide covers how to choose these based on what attackers actually do, and the endpoint security guide covers the device-level telemetry that complements network data.

One architectural note: send logs off the devices that generate them, to storage that the credentials used on those devices cannot delete. Clearing logs is a routine step for an attacker who achieves administrative access.

A note on spoofing and your outbound responsibility

RFC 2827, usually called BCP 38, describes network ingress filtering: a network should not forward traffic whose source address could not legitimately come from it. It was written to undermine denial-of-service attacks that rely on forged source addresses, and widespread adoption would remove the basis for most amplification attacks. It is worth knowing about both because your own egress filtering is a small contribution to that and because it explains why those attacks remain possible. Our guide to DDoS attacks covers the mechanics.

Suspicious link in an email or message? Protective DNS catches known-bad domains, but new ones appear constantly. Paste the link into CyberWatch AI Scan for a free check before opening it.

If you only do five things

  1. List every inbound rule from the internet and remove the ones nobody can justify.
  2. Put management interfaces in their own zone, reachable only from administrative sources.
  3. Deploy protective DNS and force all DNS through it, then keep the query logs.
  4. Separate devices that cannot be patched from everything that matters.
  5. Collect flow and authentication logs centrally and alert on internal scanning and unexplained outbound traffic.

None of that requires a large budget, and together it changes the opening scenario of this article from a company-wide incident into one compromised laptop. For the wider program this fits into, see our small business cybersecurity guide.

Frequently asked questions

What is network segmentation?

Dividing a network into separate zones with controlled traffic between them, so that compromising a device in one zone does not give an attacker reach into others. It is the control that most determines whether an intrusion stays small.

Is a firewall still useful?

Yes, in two roles. At the edge it controls what the internet can reach, which is where indiscriminate exploitation lands. Internally it enforces the boundaries between segments, which is the role most organizations underuse because their internal traffic is unrestricted.

What is the difference between IDS and IPS?

An intrusion detection system observes traffic and alerts. An intrusion prevention system sits in the traffic path and can block. The same engine often does both, and the choice is about tolerance for a false positive interrupting legitimate traffic.

What is protective DNS?

A resolver that refuses to answer lookups for known malicious domains, so a device that clicks a bad link or runs malware cannot reach its destination. It is cheap, easy to deploy and catches a meaningful share of attacks at the point of connection.

Do we still need a VPN if we use cloud services?

For reaching internal systems, yes, or something that replaces it with per-application access. What you should avoid is a VPN that drops users onto a flat internal network with broad reach, because that recreates the problem segmentation is meant to solve.

What should we monitor if we can only monitor one thing?

DNS queries and outbound connections. Nearly every intrusion needs to reach something external at some point, which makes outbound traffic the chokepoint with the best ratio of signal to effort.

Are VLANs the same as segmentation?

VLANs separate traffic logically but only provide security if routing between them is filtered. A set of VLANs that all route freely to one another is a tidier flat network, not a segmented one.

How do I segment a network that is already flat?

Start by separating categories rather than individual systems: end-user devices, servers, management interfaces, guest access and anything running unsupported software. Observe real traffic before writing rules, and begin with one high-value system rather than a comprehensive redesign.

Does encryption inside the network matter?

Yes. Internal traffic captured by an attacker who has a foothold is a common way credentials and data are harvested, and the assumption that the internal network is private has not been safe for a long time.

Sources

  1. SP 800-41 Rev. 1, Guidelines on Firewalls and Firewall Policy, US National Institute of Standards and Technology
  2. SP 800-94, Guide to Intrusion Detection and Prevention Systems (IDPS), US National Institute of Standards and Technology
  3. Layering Network Security Through Segmentation, US Cybersecurity and Infrastructure Security Agency
  4. Protective Domain Name System (DNS) Resolver, US Cybersecurity and Infrastructure Security Agency
  5. NSA and CISA Red and Blue Teams Share Top Ten Cybersecurity Misconfigurations, US Cybersecurity and Infrastructure Security Agency
  6. SP 800-92, Guide to Computer Security Log Management, US National Institute of Standards and Technology
  7. Building a Security Operations Centre (SOC), UK National Cyber Security Centre
  8. RFC 2827: Network Ingress Filtering, RFC Editor
  9. 10 Steps to Cyber Security, UK National Cyber Security Centre
Share this article
LinkedIn WhatsApp X
For organizations

Your people get these messages at work too.

CyberWatch AI sends your staff realistic practice attacks, trains the gaps it finds, and shows management exactly where the organization stands.