When an EDR agent is killed, blinded or silently unloaded, the attacker has usually already got administrator rights and is preparing the next, louder step. CISA's March 2025 advisory on Medusa ransomware describes actors who attempted to use vulnerable or signed drivers to kill or delete EDR tools and who disabled Windows Defender before deploying the encryptor. Treating the gap as a sensor fault wastes the hours you have. This template gives SOC leads, incident responders and endpoint engineers a playbook for EDR tampering and bring your own vulnerable driver (BYOVD) cases, with a decision tree for the first hour. Hardening is covered in the EDR Evasion and Defence Guide; this page is the response.
How to use this template
- Adapt to your EDR: replace [EDR console], [isolation method] and the event names with your product's equivalents. The Microsoft Defender event IDs below are examples; most EDR platforms offer comparable tamper and health signals.
- Pre-authorise containment: the SOC must be able to isolate a host or segment without waiting for a change board. Write that authority into section 4 before you need it.
- Plan for no agent: if the agent is dead, EDR isolation will not work. Each containment step names a fallback through the network or identity layer.
- Escalate deliberately: tampering on a domain controller, hypervisor or backup server moves straight to the Ransomware Playbook Template and, where domain credentials are at risk, the Active Directory Domain Compromise Response Playbook.
1. Scope and triggers
This playbook applies to all [Organisation] Windows, macOS and Linux endpoints and servers running [EDR product]. Invoke it on any of the following:
- Agent silence: a device that is online on the network (DHCP, VPN, authentication or NetFlow activity) but has stopped reporting to the EDR console beyond [15] minutes.
- Protection disabled: Defender events 5001 (real-time protection disabled), 5010 or 5012 (scanning disabled), or 5004 and 5007 (configuration changed) that do not match an approved change. Microsoft's documentation says an unexpected 5007 should be reviewed because it might be the result of malware.
- Tamper blocked: event 5013 (tamper protection blocked a change) or a tamper alert in the EDR console. A blocked attempt is still an attacker with admin rights.
- Suspicious driver load: a new kernel service or driver load (Sysmon event 6, System 7045, Security 4697) matching a known vulnerable or malicious driver, for example from the LOLDrivers catalogue, or CodeIntegrity 3076 / 3077 audit or block events for a driver.
- Exclusion abuse: new antivirus or EDR exclusions added outside the approved process.
- Service stop: security services stopped or uninstalled (ATT&CK T1489, T1685), or uninstall scripts run with administrative tools.
- Silent failure: event 3002 (real-time protection feature failed) without a following recovery event, on several hosts at once.
Out of scope: a faulty vendor update that breaks agents across the estate with no attacker involved; use the Security Tool Faulty Update Outage Playbook, but check the trigger list above before ruling out an attack.
2. Roles and contacts
| Role | Responsibilities | Named lead / deputy |
|---|---|---|
| SOC shift lead | Triage, decision tree, initial containment, incident record | [Name] / [Name] |
| Incident commander | Severity, scope expansion, ransomware escalation, executive updates | [Name] |
| Endpoint engineering | Agent reinstall, tamper protection policy, driver block rules, troubleshooting mode control | [Name] |
| Network team | Switch port, VLAN, firewall or NAC isolation when EDR isolation is unavailable | [Name] |
| Identity team | Disable accounts used for tampering, local admin password rotation, session revocation | [Name] |
| Forensics | Memory and disk capture, driver sample collection, chain of custody | [Name] or [IR retainer] |
| EDR vendor support | Agent integrity, telemetry recovery, known tampering campaigns | [Support route, contract ID] |
3. Detection and triage
First-hour decision tree
- Is the change approved? Check the change calendar and troubleshooting mode records for the device. If yes and it matches exactly, close as benign and record it. If no, continue.
- Is the device still active on the network? Look for authentication, DHCP, DNS or flow records in the last [15] minutes. If yes and the agent is silent, treat as active tampering and go to step 4.
- Is the device offline? If it is powered off or disconnected, flag it so any reconnection triggers an alert and network quarantine; continue triage on related hosts.
- Is it tier 0 or a server? Domain controllers, hypervisors, backup servers, identity and management servers go to Critical: isolate now and invoke the ransomware playbook.
- Is it more than one device? Two or more devices with tampering signals inside [24] hours means a hands-on-keyboard intrusion: declare a major incident.
- Was a driver involved? If a new kernel driver loaded before the agent stopped, treat it as BYOVD: collect the driver file and hash, and check whether the same driver exists elsewhere in the estate.
| Signal | Likely meaning | Initial severity |
|---|---|---|
| Tamper blocked (5013 or console alert), single workstation | Attacker or user with admin tried and failed | Medium: investigate the account and process |
| Protection disabled or exclusion added, no approved change | Attacker succeeded on that device | High: isolate |
| Agent silent while device active, vulnerable driver loaded | BYOVD kill of EDR | Critical |
| Tampering on several devices or any tier 0 system | Pre-ransomware staging | Critical: major incident |
Evidence to pull
- Last telemetry: the final events the agent sent, including process trees, logons and network connections in the [60] minutes before silence.
- Account used: which account ran the stop, uninstall, driver install or configuration change, and where else it has logged on.
- Driver details: service name, file path, hash, signer and load time; compare against the LOLDrivers catalogue, which lists hashes and detection content for vulnerable and malicious drivers.
- Alternative telemetry: Windows event logs forwarded to the SIEM, firewall and proxy logs, identity provider sign-ins, and NetFlow for the device.
4. Containment
- Isolate the device. Use EDR network isolation if the agent still responds. If not, use [NAC quarantine / switch port shutdown / firewall block / VPN disconnect]. Do not power it off before memory is captured unless encryption is visibly running.
- Disable the account used for tampering, revoke its sessions, and rotate the local administrator password on the device and any device sharing it (T1078.003). See the Windows LAPS Guide.
- Block the driver estate-wide. Add the driver hash to [EDR blocklist] and to an App Control driver block policy; Microsoft notes a newly applied policy only stops a running driver after a reboot, so isolate affected hosts first.
- Lock security configuration. Confirm tamper protection is on and centrally managed for the affected groups, and close any open troubleshooting mode sessions. Microsoft states Group Policy changes to tamper-protected settings are ignored, so manage these settings through [Intune / EDR console].
- Protect the management plane. Review EDR and device management console admins, API clients and recent policy changes; an attacker with console access can disable protection centrally.
- Raise monitoring on peer devices sharing the same admin accounts, subnet or software deployment path.
Decision: if any tier 0 device is affected or more than [five] devices show tampering, the incident commander decides on broader network segmentation (for example, blocking SMB and remote management between server segments) and pausing software deployment tools. Record the trade-off between business disruption and ransomware spread.
5. Eradication
- Rebuild, do not repair: a device where a malicious or vulnerable driver ran in the kernel cannot be trusted. Reimage from known-good media after forensic capture.
- Remove persistence across the scope: services and drivers created by the attacker (T1543.003), scheduled tasks, registry changes to security settings (T1112) and unauthorised exclusions.
- Close the admin path: find how the attacker got local or domain admin and fix it; most EDR tampering requires administrator rights.
- Enforce driver controls: confirm the vulnerable driver blocklist is enforced (it is on by default from the Windows 11 2022 update but enforced only with memory integrity, Smart App Control or S mode), and check CodeIntegrity Operational event 3099 shows the expected policy. Enable the attack surface reduction rule "Block abuse of exploited vulnerable signed drivers"; Microsoft notes it stops drivers being written to disk but not an existing driver loading.
- Hunt the estate for the same driver hash, service name, tools and accounts.
6. Recovery
- Return devices in rings: rebuild, enrol, confirm agent health and tamper protection status (for Defender, Get-MpComputerStatus should show IsTamperProtected as True), then reconnect.
- Verify telemetry flows end to end: generate a benign test detection and confirm it reaches the console and the SIEM.
- Watch for re-entry for [30] days with the trigger events in section 1 set to high priority for the affected scope.
- Roll back emergency blocks only through change control: driver blocks that break legitimate hardware should be replaced with a vendor-patched driver, not simply removed.
7. Communication and notification
- Internal: brief IT operations that agent reinstalls and device isolations are security actions, so they do not reverse them as support fixes.
- Executives: explain that EDR tampering is treated as a ransomware precursor and what containment has been applied.
- EDR vendor: share the driver sample, hashes and timeline; ask whether the technique is known and whether agent updates address it. Submit unknown drivers to Microsoft's driver submission process for blocklist analysis.
- External: if the case progresses to data theft or encryption, legal assesses notification under [GDPR Article 33 / sector rules]; US organisations can report ransomware to CISA or the FBI as their advisories request.
8. Post-incident review
- Timeline: privilege gained, driver dropped, agent silenced, alert raised, containment applied.
- Detection gap: how long the agent was silent before someone acted, and whether silence alerting covered that device class.
- Control gaps: memory integrity disabled, blocklist not enforced, tamper protection unmanaged, standing local admins, or exclusions inherited from old projects. See the EDR Tuning and Exclusions Management Guide.
- Threat intelligence: compare with known groups that use this technique. See the Medusa Ransomware Threat Actor Profile.
Implementation guidance
- Build silence detection first: correlate network or identity activity with missing EDR heartbeats; this single rule catches most successful kills.
- Forward key events off the endpoint: Defender 5001, 5004, 5007, 5010, 5012, 5013 and 3002, plus Sysmon 6, 7045, 4697 and CodeIntegrity 3076, 3077 and 3099, so evidence survives the agent.
- Pilot driver controls in audit mode, review audit events, then enforce in rings; have a rollback plan because Microsoft warns blocking drivers can cause malfunction or, rarely, a blue screen.
- Rehearse: run a tabletop where the EDR console shows a healthy device that is in fact silent, and test fallback isolation through the network team.
| Metric | Target example |
|---|---|
| Devices with tamper protection centrally managed and on | 100 percent of managed estate |
| Devices with driver blocklist enforced (memory integrity on or App Control policy) | Above [95] percent; 100 percent of tier 0 |
| Median time from agent silence to SOC triage | Under [15] minutes |
| Median time from triage to isolation for Critical cases | Under [30] minutes |
| Unapproved exclusions found in monthly review | Zero |
Framework mapping
| Framework | Reference |
|---|---|
| NIST CSF 2.0 | DE.CM-09, DE.AE-02, RS.MA-02, RS.MI-01, RS.MI-02, RC.RP-01, RS.CO-02, PR.PS-02 |
| NIST SP 800-53 Rev. 5 | SI-3, SI-4, SI-7, CM-7, CM-14 Signed Components, IR-4, IR-5, IR-6 |
| MITRE ATT&CK v20 | TA0112 Defense Impairment; T1685 Disable or Modify Tools; T1068; T1543.003; T1489; T1112; T1078.003 |
Standards and references
- Microsoft Learn: Microsoft recommended driver block rules
- Microsoft Learn: Protect security settings with tamper protection
- Microsoft Learn: Microsoft Defender Antivirus event IDs and error codes
- CISA, FBI and MS-ISAC: #StopRansomware: Medusa Ransomware (AA25-071A)
- NIST: SP 800-61 Rev. 3, Incident Response Recommendations
- LOLDrivers: Living Off The Land Drivers catalogue
- MITRE ATT&CK: Disable or Modify Tools
Related NHI Mgmt Group resources: EDR Evasion and Defence Guide · Top 10 EDR Evasion Techniques · EDR Guide · Ransomware Response and Recovery Guide