← Back to the cybersecurity blog

Keeping sight of DNS: a guide to resolver logging, encrypted DNS and tunnel detection

By Adrián González · CEO y socio fundador · Published: 18 November 2026 · Updated: 18 November 2026
Keeping sight of DNS

NIST now describes the recursive resolver as a security control: the place where malicious domains get blocked and where the records an investigation will need are kept. It is the thread running through SP 800-81r3, the Secure Domain Name System Deployment Guide published in March 2026 to replace the 2013 edition. As Help Net Security summarised it, NIST also wants browsers and applications with built-in encrypted DNS configured so that they cannot route around the local resolvers used for control and logging.

In the networks we review, it is often still outstanding. A browser with DNS over HTTPS (DoH) switched on can send its lookups, encrypted, to a public resolver over port 443, where they look like any other web traffic. The corporate DNS server never sees them, the SIEM holds no record of them, and the resolver's block lists stop applying to that browser. In 2026 Microsoft added DoH to the Windows Server 2025 DNS server, and Zero Trust DNS has been available on Windows 11 since late 2025; both change the options available.

This guide is aimed at infrastructure managers and SOC analysts in mid-sized organisations with Active Directory, Windows DNS servers and Microsoft Defender, with or without Microsoft Sentinel. It covers where DNS telemetry comes from, how to avoid losing it, which settings to apply and how to verify them, and which KQL hunts and Sigma rules provide a first layer of detection. Technical names were checked against each vendor's official documentation on 18 November 2026.

Where does the telemetry for a DNS lookup come from?

The path a lookup takes

An application that needs an address asks the operating system's DNS client. On Windows, the client queries the servers set on the network adapter, which on an Active Directory network are usually the domain controllers. If the name is not in an internal zone, that server forwards the query upstream, to the ISP or to a protective DNS service, which then resolves it against the authoritative servers on the internet.

Each hop leaves a different trace. The endpoint knows which process asked. The internal DNS server knows which IP address asked, but has no idea which process. The upstream resolver only sees the company's public IP, unless the service runs an agent on the device. To tie a lookup on the server back to a specific machine, SP 800-81r3 recommends correlating it with DHCP lease history.

What encrypted DNS changes

Three encrypted DNS protocols are in common use: DNS over TLS (DoT) on TCP 853, DoH on 443 (TCP and UDP) and DNS over QUIC (DoQ) on UDP 853. DoT and DoQ can be stopped with a firewall rule on 853. DoH is harder, because it shares a port with ordinary browsing and, without TLS inspection, can only be told apart by its destination. NIST suggests pairing response policy zones (RPZ) on the resolver with firewall rules to keep it in check.

Encryption on its own does not reduce visibility. Windows can use DoH with the DNS servers it is already configured with, so the query still lands on the corporate resolver: encrypted in transit, readable by whoever runs the server. Visibility is lost when an application chooses its own resolver, and browsers are the main culprits.

How does DNS tunnelling work?

The attacker relies on the fact that almost every network lets DNS lookups out to any domain. All it takes is a registered domain and control of its authoritative server. The implant encodes data in the labels of the name it looks up (each label can hold up to 63 characters) and picks up instructions from the answers, often in TXT records. Resolvers along the way normally forward the query without knowing it carries a command and control channel.

MITRE ATT&CK tracks this as T1071.004, Application Layer Protocol: DNS. Its detection strategy lists what analysts look for: frequent lookups from processes that are neither browsers nor system services, long or encoded subdomains and clients with unusual query volumes. It also notes that DNS beacons check in infrequently, which lets them blend into normal traffic.

Unit 42 described four previously undocumented DNS tunnelling campaigns in October 2024. One of them, FinHealthXDS, used twelve domains and a customised Cobalt Strike DNS beacon format against finance and healthcare targets. The research also found that domains within a campaign tend to share registration and configuration traits, which helps when clustering detections.

Which control options exist, and what does each one cover?

There are several, and no single telemetry source sees everything. The table sets out the options a mid-sized organisation can combine, with rough costs and blind spots for each:

Options for keeping DNS visibility
OptionWhat it providesRequirements and costWhat it misses
Analytic logging on the Windows DNS serverEvery query and response through the internal resolver, with source IPWindows Server 2016 or later (2012 R2 with the auditing hotfix); high event volume and SIEM ingestion costThe process behind each lookup; browser DoH traffic to public resolvers
Endpoint telemetry (Sysmon event 22 or Defender for Endpoint)The process that resolved each name; Defender's network inspectionSysmon is free but needs deployment and tuning; Defender depends on licenceDevices without an agent; lookups an application makes through its own DoH client
Blocking direct DNS egress at the firewallForces classic DNS, DoT and DoQ through the corporate resolversOutbound rules at the perimeter and in Windows Firewall; low costDoH over 443 to resolvers not on any list
DoH policies for Windows and browsersSets which resolver each browser uses and whether it encryptsGroup Policy or MDM for Edge, Chrome and Firefox; low costApplications with their own DoH outside those policies
Protective DNS (external service or in-house RPZ)Blocks malicious domains and logs every blockSubscription or list maintenance; an agent for devices that leave the networkConnections straight to IP addresses with no lookup
Zero Trust DNS on Windows 11Blocks by default outbound connections to IPs not resolved by an approved protective resolver, apart from configured exceptionsWindows 11 Enterprise or Education, E3/E5 or A3/A5 licence, protective DNS with DoH or DoTOther operating systems; needs exceptions for hard-coded IPs

Compiled by Hard2bit from Microsoft documentation, NIST SP 800-81r3 and NCSC guidance. Requirements checked on 18 November 2026.

In the networks we typically review, the best balance of cost and coverage comes from server-side logging, closed direct egress and explicit browser DoH policies, with Sysmon or Defender to attribute lookups to processes. Protective DNS adds blocking and pays off mainly where much of the workforce is remote. Zero Trust DNS is the strictest option and, because of the exceptions, probably the one that demands the most upkeep.

How is each control deployed and verified?

Analytic logging on the Windows DNS server

The Windows DNS server has two logging mechanisms that are easy to confuse. Analytic logging writes one event per query and response to the Microsoft-Windows-DNSServer/Analytical channel and is designed to stay on. Debug logging (Dns.log) is, in Microsoft's words, resource intensive and meant for temporary use only.

The analytic channel is disabled by default. It is enabled from Event Viewer (View, Show Analytic and Debug Logs; then Applications and Services Logs, Microsoft, Windows, DNS-Server, Analytical, Enable Log). Microsoft quantifies the overhead: a server handling 100,000 queries per second can lose 5% of its performance, with no visible impact up to 50,000. To check each domain controller:

# State of the DNS server analytic channel.
# Expected output includes: enabled: true.
wevtutil gl "Microsoft-Windows-DNSServer/Analytical"

In Microsoft Sentinel, the Windows DNS Events via AMA connector sends those events to the ASimDnsActivityLogs table. It needs the analytic channel enabled by hand and Azure Arc on servers outside Azure, and it can filter before ingestion by event type, record type or domain. A wildcard such as *.domain.com does not match domain.com itself, which has to be listed separately. Filters save money, but every excluded domain is one the SOC stops seeing, so exclusions should be limited to high-volume internal names and documented. Once the connector is live, a short hunt confirms that each domain controller is reporting:

// Events received per server and type over the last day.
ASimDnsActivityLogs
| where TimeGenerated > ago(1d)
| summarize events = count() by Dvc, EventOriginalType

Closing direct DNS egress

The cheapest control is an outbound rule on the perimeter firewall that lets only the corporate resolvers reach the outside world on ports 53 (UDP and TCP) and 853 (TCP and UDP), over both IPv4 and IPv6. The NCSC's protective DNS guidance recommends this to stop clients bypassing the approved resolver.

Before blocking, an inventory is needed of who uses that path today: mail servers that query block lists through their own resolver, DMZ servers, devices with 8.8.8.8 hard-coded and some VPN clients. Running the rule in log-only mode for a week or two surfaces those exceptions without breaking anything. The final test runs from a user workstation:

# Should time out if direct egress is closed.
Resolve-DnsName www.example.com -Server 1.1.1.1 -DnsOnly
# Should succeed against the corporate resolver.
Resolve-DnsName www.example.com -DnsOnly

An answer to the first command means a path is still open. The test should be repeated from the guest network, over VPN and at each site, because rules tend to differ from one location to the next.

The Windows DNS client DoH policy

Windows 11 and Windows Server 2022 or later include the Configure DNS over HTTPS (DoH) name resolution policy under Computer Configuration, Administrative Templates, Network, DNS Client. It has three settings. With Allow DoH, Windows encrypts only when the configured servers support it; with Prohibit DoH, it never encrypts; with Require DoH, resolution fails if the server cannot do DoH.

The CIS Windows 11 benchmark asks for Allow DoH or higher, which is compatible with visibility for the reason given above. Microsoft advises against Require DoH on domain-joined machines because Active Directory depends on DNS; even now that the Windows Server 2025 DNS server accepts DoH, requiring it means every domain controller needs a valid certificate and its DoH template registered on every client (Add-DnsClientDohServerAddress). To review the DoH servers Windows knows about:

# DoH templates known to the Windows DNS client.
# A template is only used if that IP is one of the adapter's DNS servers.
Get-DnsClientDohServerAddress

Managed browsers

Browsers ship with a DNS client and DoH settings of their own. Edge and Chrome use the DnsOverHttpsMode policy, which takes three values: off disables DoH, automatic uses it where possible and falls back to classic DNS on failure, and secure always encrypts and never falls back. The Edge documentation states that a managed device with the policy unset sends no DoH queries. For Chrome, the Chromium source turns DoH off by default on desktop when machine-level policies exist or the device is domain-joined.

Leaning on that implicit behaviour is fragile, since a laptop with no machine policies that is not domain-joined changes the outcome. The policy should be set explicitly: off where the corporate resolvers do not speak DoH, and secure with DnsOverHttpsTemplates pointing at an in-house resolver that does.

In secure mode the browser cannot resolve names when it cannot reach that server, so a laptop off the corporate network with no route to it loses browsing altogether. The CIS level 2 benchmark for Chrome recommends secure, which only fits with visibility when the template points at a resolver the organisation controls.

# Edge and Chrome registry values (REG_SZ).
# HKLM\SOFTWARE\Policies\Microsoft\Edge DnsOverHttpsMode = off
# HKLM\SOFTWARE\Policies\Google\Chrome DnsOverHttpsMode = off
# edge://policy and chrome://policy should list it as a mandatory machine policy.
# On Windows, gpresult /h report.html shows which GPO applies it.

Firefox is managed through the DNSOverHTTPS policy, set in policies.json or through Group Policy (Software\Policies\Mozilla\Firefox\DNSOverHTTPS). Its keys are Enabled, ProviderURL, Locked, ExcludedDomains and Fallback. For most organisations, Enabled set to false and Locked set to true will do, and about:policies confirms what the browser has picked up.

{ "policies": { "DNSOverHTTPS": { "Enabled": false, "Locked": true } } }

Mozilla also provides a signalling domain, use-application-dns.net. When the network's resolver gives a negative answer for it (NXDOMAIN, SERVFAIL or a reply with no A or AAAA records), Firefox switches off DoH that was on by default. Mozilla's page is clear that the signal does not override a user who enabled DoH manually, so it backs up the policy without replacing it. Firefox does not enable DoH by default in Spain, so the signal has little practical effect there.

# From a workstation: if a zone or rule for that name has been created on the resolver, it should return NXDOMAIN.
Resolve-DnsName use-application-dns.net

Does the Windows Server 2025 DNS server support DoH?

Yes, since the June 2026 update, although only the client-to-server leg is encrypted. Before then the Windows DNS server did not offer DoH as a generally available feature, so encrypting DNS while keeping an in-house resolver meant running non-Microsoft software, commercial or open source.

Since the June 2026 security update (KB5094125), Microsoft lists DoH as generally available on the Windows Server 2025 DNS server. According to Microsoft's documentation, only the client-to-server leg is protected. Traffic to forwarders and authoritative servers, zone transfers and, by default, dynamic updates remain unencrypted, and DNS policies that filter on transport protocol do not match DoH queries.

Protective DNS

A protective DNS service checks each query and response against threat intelligence lists of malicious domains and blocks matches. It is wired in by pointing the internal DNS servers' forwarders at the service, or by running an in-house RPZ. The NCSC recommends sending the service's block logs to the SIEM, and SP 800-81r3 describes cloud, on-premises and hybrid designs, favouring hybrid. Laptops working off-network need an agent on the device, otherwise they resolve through whatever DNS the home router hands out.

Zero Trust DNS on Windows 11

Zero Trust DNS (ZTDNS) ties the Windows DNS client into the Windows Filtering Platform and blocks all outbound IPv4 and IPv6 traffic by default. The exceptions are the configured protective DNS servers, explicitly allowed ranges and discovery traffic (DHCP, DHCPv6 and NDP). Each answer from the protective resolver opens a time-limited exception for the IPs it returned. It requires Windows 11 Enterprise or Education and a protective resolver that supports DoH or DoT. Microsoft announced it as generally available in November 2025 (Ignite 2025), following a public preview earlier that year.

The deployment guide starts in audit mode, which logs what would have been blocked without blocking it. Applications with hard-coded IPs need manual exceptions, and the guide shows how to add them for Microsoft 365 and Teams WebRTC traffic, which bypasses DNS; its subnets are examples and must be replaced with Microsoft's official Microsoft 365 address list. The same guide disables BuiltInDnsClientEnabled in Edge and Chrome so that browsers resolve through the operating system, but that policy does not control DoH, so DnsOverHttpsMode also has to be set to off, or to secure with the protective resolver's template.

Switching to enforcement (audit=no) cuts everything without an exception, and rollback requires local administrator access, so the pilot belongs on devices with physical access, with the rollback command to hand:

# Enable ZTDNS in audit mode and review the applied configuration.
netsh ztdns set state enable=yes audit=yes
netsh ztdns show settings
# Emergency rollback if enforcement cuts off a service.
netsh ztdns set state enable=no audit=no

Detection rules and how to test them safely

The hunts below are a starting point and should be tuned against local data. The first uses the normalised _Im_Dns parser from the ASIM DNS schema over Windows DNS server events; with other normalised sources the event-type filter changes. The next two run in Microsoft Defender XDR advanced hunting. Our guide to detection engineering sets out a method for measuring what they cover before they go live.

Possible tunnel: many long, unique subdomains

// Possible DNS tunnel in the last hour, over Windows DNS server events.
// In that source, 256 is a query received; recursion events carry the server itself as source.
// First check the values: summarize count() by EventOriginalType.
let unique_threshold = 200;
let length_threshold = 40;
let internal_suffix = "corp.example";
_Im_Dns(starttime=ago(1h), endtime=now())
| where EventOriginalType == "256"
// Lower case, no trailing dot, no reverse zones, SRV records or internal zones.
| extend q = trim_end(@"\.", tolower(DnsQuery))
| where isnotempty(q) and not(q endswith ".arpa") and not(q startswith "_") and not(q endswith strcat(".", internal_suffix) or q == internal_suffix)
| extend labels = split(q, ".")
| where array_length(labels) >= 3

The same hunt carries on with the parent domain and the aggregation; paste both parts together:

// The parent domain is approximated by the last two labels.
| extend parent = strcat(tostring(labels[(-2)]), ".", tostring(labels[(-1)]))
| extend sub = substring(q, 0, strlen(q) - strlen(parent) - 1)
// Distinct subdomains and average length per source and parent domain.
| summarize unique_subs = dcount(sub), avg_length = avg(strlen(sub)), types = make_set(DnsQueryTypeName, 5) by SrcIpAddr, parent
| where unique_subs > unique_threshold and avg_length > length_threshold
| order by unique_subs desc

Most false positives come from well-known services: mail servers querying DNS block lists, antivirus agents checking reputation with hashes in the subdomain, vendor telemetry and content delivery networks that use random hostnames. The sensible approach is to start with high thresholds, review a week of results and keep an exclusion list with a date and a reason for every entry. The two-label shortcut breaks on suffixes such as co.uk or com.es, which would lump unrelated domains together.

The false negatives are just as predictable. A slow beacon that checks in a few times an hour stays under the threshold, and a device sending DoH to an outside resolver never reaches the internal server. A 24-hour window with different thresholds helps with the first; the next hunt deals with the second.

DoH or DoT outside policy

// Connections to public resolvers over DoT (853) or DoH (443).
let doh = dynamic(["dns.google", "cloudflare-dns.com", "mozilla.cloudflare-dns.com", "chrome.cloudflare-dns.com", "1dot1dot1dot1.cloudflare-dns.com", "dns.quad9.net", "doh.opendns.com", "dns.nextdns.io"]);
// Default Windows DoH list, IPv4 and IPv6.
let doh_ips = dynamic(["1.1.1.1", "1.0.0.1", "2606:4700:4700::1111", "2606:4700:4700::1001", "8.8.8.8", "8.8.4.4", "2001:4860:4860::8888", "2001:4860:4860::8844", "9.9.9.9", "149.112.112.112", "2620:fe::fe", "2620:fe::fe:9"]);
DeviceNetworkEvents
| where Timestamp > ago(7d)
| where ActionType in ("ConnectionSuccess", "ConnectionAttempt", "ConnectionFailed")
| where RemotePort == 853 or (RemotePort == 443 and (RemoteUrl has_any (doh) or RemoteIP in (doh_ips)))
// Grouped by device and process, with the destination.
| summarize connections = count(), first_seen = min(Timestamp), last_seen = max(Timestamp) by DeviceName, InitiatingProcessFileName, ActionType, RemoteUrl, RemoteIP, RemotePort
| order by connections desc

A browser in the results usually means a device the policy has not reached. Anything else needs a closer look. Developer tools, VPN clients and security agents may use DoH legitimately, but so does malware trying to avoid the corporate resolver. Once egress is closed, failed attempts are what matter, which is why the hunt includes them. The list of domains and IPs is incomplete by design: there are hundreds of public DoH resolvers and an attacker can run their own.

TXT requests to unconfigured servers

In 2023 Microsoft published a hunting example built on Defender for Endpoint's DnsConnectionInspected action type, which pulls the query, its type and the answers out of network traffic. The adapted version below looks for answered outbound TXT requests sent to servers other than the device's configured DNS servers, and leaves out the DNS servers themselves, which recurse to the internet as a matter of course. The action type should first be confirmed in the tenant, with DeviceNetworkEvents | where ActionType == "DnsConnectionInspected" | take 10.

// DNS servers configured on devices over the last day.
let known_dns = DeviceNetworkInfo
| where Timestamp > ago(1d)
| mv-expand dns = todynamic(DnsAddresses)
| extend dns = tostring(dns)
| where dns !in ("", "::", "127.0.0.1", "0.0.0.0")
| distinct dns;

Still in the same hunt, the resolvers are excluded and the TXT requests filtered:

// Domain controllers and other resolvers, which query internet servers.
let dns_servers = dynamic(["dc01.corp.example", "dc02.corp.example"]);
DeviceNetworkEvents
| where Timestamp > ago(1d)
| where ActionType == "DnsConnectionInspected"
| where DeviceName !in~ (dns_servers)
| extend af = todynamic(AdditionalFields)
| extend direction = tostring(af.direction), qtype = tostring(af.qtype_name), query = tostring(af.query), answers = todynamic(af.answers)
// Answered outbound TXT requests to any other server.
| where direction == "Out" and qtype == "TXT" and RemoteIP !in (known_dns) and array_length(answers) > 0
| project Timestamp, DeviceName, InitiatingProcessFileName, RemoteIP, query

With direct egress closed and the resolvers excluded, this hunt should return few rows. Each one is a device that talked to a DNS server it should not be using, and the initiating process usually shows whether it is an admin tool or something that has no business being there.

Sigma rule for Sysmon

On devices without Defender, Sysmon event 22 records processes' DNS lookups (those that go through the Windows DNS client), successful or not, cached or not, from Windows 8.1 onwards. Several widely used Sysmon configurations filter this event for volume, so collection should be confirmed. The rule flags non-browser processes looking up public DoH endpoints, a common step before using DoH directly:

title: Non-browser process looks up a public DoH endpoint
id: 475cda16-91b2-4f3d-9111-27a51bb8cb6e
status: experimental
description: Detects non-browser processes looking up public DoH resolver domains, a possible attempt to bypass corporate DNS.
author: Hard2bit
date: 2026-11-18
tags: [attack.command-and-control, attack.t1071.004, attack.t1572]
logsource: {product: windows, category: dns_query}
# Browsers are excluded by file name.
detection: {selection: {QueryName: [dns.google, cloudflare-dns.com, mozilla.cloudflare-dns.com, chrome.cloudflare-dns.com, 1dot1dot1dot1.cloudflare-dns.com, dns.quad9.net, doh.opendns.com, dns.nextdns.io]}, filter_browsers: {Image|endswith: ['\chrome.exe', '\msedge.exe', '\msedgewebview2.exe', '\firefox.exe', '\brave.exe']}, condition: selection and not filter_browsers}
falsepositives: [VPN clients and security agents with legitimate DoH, Developer and network troubleshooting tools]
level: medium

The rule misses names an application resolves through its built-in DoH client; event 22 depends on the Windows DNS client, although Microsoft does not document this explicitly. It also misses connections made straight to a resolver's IP, and the file-name filter can be dodged by renaming a binary, so filtering on the full path under Program Files makes it harder to evade. It is more useful alongside the hunt on ports 443 and 853 than on its own. sigma check validates the syntax before conversion.

Testing safely

Validation needs no offensive tooling. From a test machine covered by written authorisation, more than 200 random subdomains longer than 40 characters are resolved under a domain the organisation owns, outside the internal zone and with no encoded data. The first KQL hunt should pick them up in the next window.

On the DoH side, a lab browser with no policy and DoH switched on by hand should appear in the second hunt, and a DoH attempt to 1.1.1.1 on 443 should fail if the block covers those IPs. If ZTDNS is in audit mode, its logs will show what it would have cut. Each test is announced to the SOC beforehand and recorded with date, device and result, so that it doubles as audit evidence.

Common mistakes and the limits of the control

  • Require DoH pushed through policy to domain-joined machines, for the reasons covered above.
  • DNS server debug logging left on permanently instead of the analytic channel.
  • An assumption that Chrome or Edge will switch DoH off by themselves on managed devices, with no policy set.
  • Whole domains excluded from the connector to save on ingestion, with no recorded reason and no review.
  • Remote laptops counted as covered while they resolve through the home router's DNS.
  • A 53/853 block rolled out with no prior inventory, which takes down mail servers, DMZ hosts or devices with hard-coded DNS.

Some gaps cannot be closed with DNS settings at all. Malware that connects to IP addresses written into its code generates no lookups, and malware that tunnels over HTTPS to legitimate services has no need for suspicious DNS. DNS is a valuable detection source because nearly everything touches it, yet it needs network and endpoint telemetry to cover those gaps, as we explain in our piece on EDR blind spots. Internal network segmentation rounds off the egress picture: a device that can only reach the internet through controlled services leaves fewer routes for data exfiltration.

Evidence for ENS, ISO 27001 and NIS2

The controls in this guide produce concrete audit evidence. The table maps them to measures in Spain's National Security Framework (ENS, Royal Decree 311/2022) and to ISO 27001:2022 Annex A controls:

Mapping to compliance frameworks
ControlENSISO 27001:2022Evidence
DNS server analytic logging sent to the SIEMop.exp.8 Activity logging (Registro de la actividad); op.mon.3 Monitoring (Vigilancia)8.15 Logging; 8.16 Monitoring activitieswevtutil output per server; connector configuration and retention
Closed direct DNS egressmp.com.1 Secure perimeter8.20 Networks securityFirewall rule export; dated Resolve-DnsName test
DoH policies for Windows and browsersop.exp.2 Security configuration8.9 Configuration managementGroup Policy or MDM report; screenshots of edge://policy and about:policies
Protective DNSop.exp.6 Protection against malicious code; op.mon.1 Intrusion detection8.23 Web filtering; 8.7 Protection against malwareService contract and configuration; block logs in the SIEM
Detection queriesop.mon.1 Intrusion detection; op.mon.3 Monitoring (Vigilancia)8.16 Monitoring activitiesVersion-controlled rules; validation results

ENS measures apply according to the system's category.

Implementing Regulation (EU) 2024/2690 fleshes out Article 21 of NIS2 for digital providers such as DNS services, cloud, data centres and managed services, and point 3.2.3 lists inbound and outbound network traffic among the events to be logged. For other essential and important entities it is a fair guide to what a supervisor will expect. ENS and ISO 27001 ask the same questions in different words: what is logged, for how long, and who reviews it.

Where should work start?

By closing direct DNS egress at the firewall and setting an explicit DoH policy in browsers. These two low-cost changes route the vast majority of office lookups through the corporate resolver. Without them, server logging cannot record queries that never arrive and protective DNS cannot block what it is never asked. With them, analytic logging and the hunts in this guide work on near-complete data for the internal network, and DoH to unknown resolvers becomes the gap left to watch.

What follows depends on resources. A team with a managed SOC or its own threat hunting capability will get more out of endpoint telemetry and regular tunnel hunts. An organisation with a large mobile workforce gains more from protective DNS with an agent. For what else belongs in the SIEM, there is our guide to which logs to prioritise and how long to keep them.

And if a hunt does uncover a tunnel, the first thing an incident response team will ask for is the resolver log, to work out how long it has been running and which machines are involved. The log only exists if someone switched it on beforehand.

Sources and methodology: NIST SP 800-81r3; official documentation from Microsoft (Windows DNS server and client, Zero Trust DNS, Edge, Defender for Endpoint and Microsoft Sentinel), Google Chrome and Mozilla; the NCSC's protective DNS guidance; MITRE ATT&CK; Unit 42 research and the Help Net Security summary. Technical names and feature status checked as of 18 November 2026.
Disclaimer: this guide is for general information. The settings and detection rules are offered as a starting point and should be tested in a non-production environment and adapted to each organisation before use in production. Technical names may change in later product versions.

Frequently asked questions

Is DNS over HTTPS bad for enterprise security? ▾

Not in itself: it encrypts lookups so third parties cannot read or tamper with them. The trouble starts when an application sends it to an outside resolver and cuts out the corporate one, which is where logging and blocking happen. With an in-house resolver that accepts DoH, such as the Windows Server 2025 DNS server since June 2026, encryption is possible without losing visibility.

What is the difference between analytic and debug logging on a Windows DNS server? ▾

Analytic logging writes one event per query to an event channel and is meant to stay on; Microsoft estimates a 5% overhead at 100,000 queries per second. Debug logging records traffic in great detail to a text file and is meant only for short-term troubleshooting.

How can an organisation tell whether its browsers send DoH to public resolvers? ▾

A sample of devices checked with edge://policy, chrome://policy and about:policies shows whether the policy is applied. For the full picture, endpoint telemetry should be searched for connections on ports 853 and 443 to known DoH resolver domains, grouped by process.

Is blocking outbound port 53 enough? ▾

It covers classic DNS but not DoT or DoQ, which use 853, nor DoH, which runs over 443. Port 853 also has to be closed, DoH managed through Windows and browser policies, and connections to public resolvers monitored to catch whatever slips through.

What false positives does DNS tunnel detection produce? ▾

Almost always legitimate services that pack information into subdomains, such as the block lists mail servers query. They are handled with a dated, reasoned exclusion list and thresholds set after a week of observed traffic.

Does Zero Trust DNS replace the firewall? ▾

No. It is a control on the Windows 11 Enterprise or Education device itself that only allows connections to addresses resolved by an approved protective resolver, apart from explicitly allowed ranges and discovery traffic. The perimeter is still needed for everything that is not Windows 11 Enterprise or Education: servers, printers, phones and guest devices.

Which evidence from this guide helps in an ENS or ISO 27001 audit? ▾

wevtutil output showing the analytic channel is on, the firewall rule export, the browser policy report and version-controlled detection rules with dated test results. They support ENS op.exp.8, op.exp.2, op.mon.1, op.mon.3 and mp.com.1, depending on system category, and ISO 27001:2022 controls 8.15, 8.16, 8.9 and 8.20.

Want a straight answer on scope, priorities and price?

Most organizations reach us mid-question: something needs fixing, nobody has scoped it, and finance wants a number. Thirty minutes with a technical consultant — not a salesperson — gets you a defined scope, priorities ranked by risk and a price range. With what comes out of that call, we turn it into a fixed proposal. We work with organizations in Spain, across the EU and in LATAM.

If your situation is different, tell us anyway — we also take one-off questions on cybersecurity and regulatory compliance.

Based in Spain · Working across the EU and LATAM · ENS High · ISO 27001 · We usually reply in under 24 business hours