Canadian data residency · made, hosted, and operated in Canada

Early access — 90 days →

MITRE ATT&CK Enterprise · TA0112

Defense Impairment: how an intruder breaks your ability to see

Defense ImpairmentTA0112

An intruder who cannot be seen does not have to be fast. This is the tactic that covers the sabotage of your monitoring itself — what it costs, what it looks like from the outside, and what still survives it.

Codein Cybersecurity November 22, 2026 8 min read

For the owner

Some intrusions do not try to be invisible. They try to make you unable to tell what happened.

Logging is turned off. A firewall rule is deleted. A certificate is installed that makes a machine trust the intruder. The second factor is switched off for a service account. Nothing is stolen on that day, and from the outside the company looks entirely normal. But the organisation has quietly lost the ability to see what is happening on its own systems, and it usually finds out at the worst possible moment — during an unrelated incident, a regulator's question, or a customer's demand for proof.

The business cost of this is different from the cost of a direct theft, and it is easy to underestimate. A company that cannot reconstruct what happened cannot answer its insurer, cannot meet its reporting obligations, cannot convict anyone, and cannot tell whether it has been breached once or a hundred times. It also cannot tell when the tampering started, which is usually the question that decides the size of the answer.

Three questions are worth asking your own team:

  1. If our logging stopped on one machine tomorrow, would we know? A reporting agent that can be stopped by the same credentials used to work on the machine is not a safety net.
  2. Do we know what our systems are supposed to be configured like? Not what they should be according to a policy document — what they actually were on Monday, so that a change on Tuesday has something to compare against.
  3. Who is allowed to turn off logging, weaken authentication or install trust certificates, and how would we know if they did? This is often a very short list of people, and the list is not always the list of people whose names appear on the change request.

None of this is a large purchase. It is mostly the discipline of recording what the configuration was, and of reporting the changes from a place the person making them does not control.

For the analyst

The tactic

MITRE ATT&CK files Defense Impairment under TA0112. It is the newest tactic in the Enterprise matrix, and it exists because the older ones did not have a home for it. Its subject is not the intrusion's goal but the intrusion's effect on the defender: the mechanisms, pipelines and tooling that would have reported the intrusion are degraded, disabled, or turned into the attacker's tool.

Do not confuse it with Stealth (TA0005), which is the tactic for hiding one action in plain sight, and not with the older defense evasion framing of the same idea. Stealth hides an event; Defense Impairment removes the recorder. The distinction matters operationally: a stealth detection can be built on your existing event stream, whereas a defense-impairment detection has to survive the impairment of that stream.

The matrix files 57 techniques under it, and the pinned SigmaHQ corpus carries 362 rules that name techniques in this tactic.

The techniques the corpus detects

TechniqueNameRules in the corpus
T1685Disable or Modify Tools161
T1112Modify Registry81
T1685.001Disable or Modify Tools: Disable or Modify Windows Event Log26
T1686.003Disable or Modify System Firewall: Windows Host Firewall19
T1556Modify Authentication Process12
T1553.004Subvert Trust Controls: Install Root Certificate10
T1685.005Disable or Modify Tools: Clear Windows Event Logs7
T1686Disable or Modify System Firewall7
T1484.001Domain or Tenant Policy Modification: Group Policy Modification6
T1553.005Subvert Trust Controls: Mark-of-the-Web Bypass6
T1686.001Disable or Modify System Firewall: Cloud Firewall6
T1222.002File and Directory Permissions Modification: Linux and Mac Permissions4
T1685.006Disable or Modify Tools: Clear Linux or Mac System Logs4
T1222.001File and Directory Permissions Modification: Windows Permissions3
T1556.002Modify Authentication Process: Password Filter DLL3
T1556.006Modify Authentication Process: Multi-Factor Authentication3
T1685.002Disable or Modify Tools: Disable or Modify Cloud Log3
T1207Rogue Domain Controller2
T1222File and Directory Permissions Modification2
T1553Subvert Trust Controls2
T1686.002Disable or Modify System Firewall: Network Device Firewall2
T1484Domain or Tenant Policy Modification1
T1484.002Domain or Tenant Policy Modification: Trust Modification1
T1553.001Subvert Trust Controls: Gatekeeper Bypass1
T1553.002Subvert Trust Controls: Code Signing1
T1553.003Subvert Trust Controls: SIP and Trust Provider Hijacking1
T1556.004Modify Authentication Process: Network Device Authentication1
T1578Modify Cloud Compute Infrastructure1
T1578.003Modify Cloud Compute Infrastructure: Delete Cloud Instance1
T1599.001Network Boundary Bridging: Network Address Translation Traversal1
T1685.004Disable or Modify Tools: Disable or Modify Linux Audit System Log1
T1689Downgrade Attack1
T1690Prevent Command History Logging1

Six ways a defence gets impaired

Disabling the recorder (T1685, T1685.001, T1685.005, T1685.006, T1685.004, T1685.002). Turning off the Windows event log, clearing it, clearing Linux or macOS system logs, disabling the Linux audit system log, or turning off a cloud log service. In practice this is a service being stopped, a configuration value being written, or a command that resets a policy. The telemetry to reason over is a windows_sysmon event whose event_type is a service or configuration change, the command_line behind it, the file_path of the tool that made the change, and the user_name that made it — plus the linux_auditd execution record on the Linux side, where the arguments carry the operation. The detection has to be for the operation, not for the tool: auditctl and the Windows event log utilities are ordinary administration work.

Turning off the gate (T1686, T1686.003, T1686.001). Disabling the Windows host firewall, deleting a cloud firewall rule, or changing a network device's policy. Firewall changes show in firewall_syslog only for the traffic that follows, which is the problem: a disabled firewall appears as an absence. The reliable signal is the change itself — the configuration API call, the administrative command, the rule deletion recorded by the cloud provider's own audit trail.

Editing the record (T1112, T1222, T1222.001, T1222.002). Registry modification, and changes to file and directory permissions on Windows and on Linux or macOS. Permission changes are the more interesting half: making a directory writable so a loader picks up something the attacker placed there, or making a log file writable so it can be truncated, changes the meaning of a path rather than its contents. In the normalised columns this is file_path plus the command_line of the permission call.

Weakening the trust (T1553, T1553.001, T1553.002, T1553.003, T1553.004, T1553.005). Installing a root certificate, bypassing Gatekeeper, hijacking a trust provider, subverting code signing, or defeating the mark-of-the-web so a downloaded file stops being treated as downloaded. The consequence is that the operating system's assurance becomes the attacker's: after a root certificate is installed, the machine considers the attacker's traffic legitimate. The technical change is small; the consequence is total.

Changing what counts as authenticated (T1556, T1556.002, T1556.004, T1556.006). Modifying the authentication process — a password filter library, network device authentication, or the state of multi-factor authentication for a set of users. Here the detection surface is the identity system rather than the endpoint: keycloak, m365_audit and gworkspace audit records carry event_type, user_name, src_ip and message.

Changing policy and the estate (T1484, T1484.001, T1484.002, T1207, T1578, T1578.003, T1599.001, T1689, T1690). Group policy and trust modification in the directory, a rogue domain controller, changes to cloud compute infrastructure including the deletion of a cloud instance, network boundary bridging, a downgrade of a security setting, and the prevention of command history logging. These change the shape of the estate rather than the state of one machine.

Where the telemetry comes from

  • windows_sysmon — service and configuration changes, registry writes, process creation with the command line that made the change.
  • linux_auditd — execution records carrying the arguments of the change, and configuration writes to the audit system.
  • keycloak, m365_audit, gworkspace, cloud — changes to authentication policy, role assignments, log services, firewall rules and compute resources. For this tactic the identity and cloud audit trails are not a supplement; they are often the primary source.
  • firewall_syslog — the traffic consequence of a firewall change, which is the slowest but most certain signal of all.

The structural difficulty of this tactic is that most of these changes are recorded by the subsystem being changed: a log service that has been turned off does not report that it was turned off. That is why the collection path must be independent of the thing being collected, and why a detector that reads only the impaired subsystem has a known failure mode.

What the numbers do and do not tell you

362 rules is a substantial block of the corpus, and it is concentrated: Disable or Modify Tools is the largest single technique in the tactic, with the Windows event log and the firewall families behind it. The distribution is informative rather than reassuring. The Windows endpoint families are well covered because their artefacts are visible in ordinary telemetry; the directory and identity families are thinly covered because their artefacts live in vendor audit trails that many organisations do not collect at all. The takeaway is not that the corpus is weak, but that coverage for this tactic is a property of your collection, not of the rule set. A perfectly good rule that reads a log you do not ship is not a detection.

Detection engineering notes

  • The alert must arrive by a path the attacker has not touched. This is the one rule that matters most in this tactic. If the collector, the configuration channel and the alert path all depend on the same administrative credential and the same subsystem, they fail together.
  • Compare against a recorded baseline, not against a policy document. "The firewall is off" is weak; "the firewall was on yesterday and is off now" is strong. Keep a known-good configuration snapshot per host.
  • Alert on the pairing. A recorder being disabled and a payload being installed usually happen minutes apart, on the same host, by the same account. The correlation is the detection; the isolated event is only the first half.
  • Treat trust changes as high severity regardless of context. A new root certificate, a code-signing change, a trust modification in the directory: these are never routine, and they are cheap to detect because they are rare.
  • Tune with change windows, not with permanent exceptions. Log rotation, agent upgrades, firewall edits during a planned change and multi-factor rollouts all touch these settings. A rule that ignores the change calendar will be muted by the first honest maintenance night.

Where analysts get this wrong, and what a false negative looks like

The most common error is assuming that "we collect the logs" is an answer to "can the logs be collected?". The second is treating the change record as the detection, when the change record was written by the subsystem that was impaired. The third is forgetting that most of these changes are made with an administrator's own credentials, which makes them look legitimate in the very record you would search.

The false negative to hold in mind is a two-part event. An intruder disables logging on one host and does the rest of the work through the directory, whose logs travel the same tooling. The endpoint goes quiet and the directory's records are partial. The surviving evidence is in the identity provider's own trail, which nobody was watching, and the one configuration change that tells the whole story is a legitimate-looking entry made with a valid administrator's credentials. The second part is worse than the first: a malicious or coerced insider makes the change through the normal path, so the change record contains their own name and proves nothing.

Where DetectMonkey fits

A defence that can be impaired is not a defence, and the question for a Canadian owner is not only whether monitoring exists but where its records live, who can reach them, and whether they survive an attack on the systems they describe. That is a residency question as much as a technical one: when the evidence about your estate sits on private servers operated by Codein in New Brunswick, Canada, it is not reachable by a supplier in another jurisdiction on a foreign government's request, and it is not lost when the systems it describes are re-imaged in the middle of an incident.

DetectMonkey is built on that footing. The collectors report endpoint, identity and firewall telemetry to a private pipeline that is evaluated as the events arrive, the pinned corpus of 3164 detection rules covers the configuration and tool changes described above, and every alert carries its evidence in English and French. The platform detects and alerts. It does not isolate, quarantine, block or remediate anything on your systems, and it does not change your configuration. Telemetry is processed and stored on private servers operated by Codein in New Brunswick, Canada — not in a public cloud.

Early access is 90 days at no charge, for up to 100 seats, with founder-led onboarding and no card required.

See it on your own telemetry

Early access is 90 days at no charge, for up to 100 seats, founder-led onboarding, no card required.