Eradication Validation Checklist
How do you know you're actually clean? Eradication that ends when you stop looking is not eradication — it's wishful thinking with a closed ticket attached. This chapter builds the systematic methodology for declaring eradication complete with confidence rather than hope.
The IR team spends four days investigating a ransomware precursor intrusion. They find Cobalt Strike beacons on fourteen workstations, identify the initial access vector (a phishing email), and remove all persistence mechanisms they find. On day five, they declare eradication complete and close the incident ticket. Six days later, the attacker is back — this time deploying ransomware. The investigation reveals a scheduled task on a domain controller that had been missed entirely. The scope of the original eradication effort was defined as "workstations only" because the forensics team assumed the DC hadn't been touched. The attacker had spent two weeks on the DC before the initial detection, and nobody checked.
The lesson is not "check domain controllers" — that's too specific. The lesson is that eradication scope defined by assumption is not eradication scope. You check everything the attacker could have touched, systematically, using a framework, not intuition.
What "Clean" Actually Means
The word "clean" in incident response has a precise meaning that is almost never stated explicitly. In practice, teams declare eradication when they have removed the malware and persistence they found. That is not the same as clean.
Clean means: every persistence mechanism from every persistence class has been systematically verified as absent on every host within the defined scope. It is a negative assertion — not "we found and removed the persistence" but "we checked every place persistence could exist and confirmed none is present." The distinction matters because attackers routinely plant secondary persistence mechanisms specifically because they know IR teams stop after finding the first.
The Eradication Completeness Gap
═══════════════════════════════════════════════════════════════════
What teams typically do:
─────────────────────────────────────────────────────────────────
1. Find the malware that triggered the alert
2. Find the persistence mechanism that loaded the malware
3. Remove both
4. Declare eradication complete
What eradication actually requires:
─────────────────────────────────────────────────────────────────
1. Define scope: every host the attacker could have touched
(not: every host where we found obvious compromise)
2. For every host in scope:
→ Check every persistence class (8+ classes on Windows alone)
→ Verify authentication baseline returned to pre-incident state
→ Verify no new accounts, no modified privileges
→ Verify network behavior normalized
3. For the AD environment:
→ Check for new Group Policy Objects
→ Check for modified GPOs
→ Check for new admin accounts
→ Check KRBTGT key rotation status
→ Check for new scheduled tasks at domain level
4. Document: what was checked, by whom, when, result
5. Sign-off: named individual takes responsibility for clean declaration
The gap between steps 1-4 above and steps 1-5 of the full
methodology is where most reinfections originate.
Defining Eradication Scope
Before you can validate eradication, you must define what you're validating. Scope definition is an output of the investigation phase — specifically, the blast radius analysis. Every host the attacker authenticated to, executed code on, or connected to from a compromised host is in scope. Every host is in scope until evidence clears it, not until assumption clears it.
| Evidence type | Clears a host from scope? | Reasoning |
|---|---|---|
| No EDR alerts on the host | No | Attacker may have operated without triggering detection; EDR gaps exist |
| Host not in lateral movement log analysis | Only if log analysis is complete and logs cover the full incident window | Log gaps (no Sysmon, no process creation logging) mean absence of evidence ≠ evidence of absence |
| No active beaconing from host at time of detection | No | Attacker may have been on the host earlier and established persistence that isn't actively beaconing |
| Positive evidence of no attacker access in complete logs | Yes, provisionally | If logs are complete, cover the full incident window, and show no authentication or execution from attacker-controlled accounts — this is reasonable |
| Host was offline during the entire incident window | Yes | Attacker cannot have reached a host that was not networked |
Scope should be defined by potential access, not confirmed access. If the attacker compromised a host with domain admin credentials, every host in the domain is in scope for at least a check — because domain admin can reach every host. Limiting scope to "hosts with EDR alerts" or "hosts we found malware on" guarantees you miss the hosts the attacker was more careful on.
Re-imaging vs Cleaning in Place
One of the most consequential decisions in eradication is whether to re-image compromised hosts or clean them in place. Neither approach is universally correct. The choice depends on confidence level, operational constraints, and the nature of the compromise.
Re-image vs Clean in Place Decision Framework
═══════════════════════════════════════════════════════════════════
RE-IMAGE when:
─────────────────────────────────────────────────────────────────
• You lack confidence in the completeness of your forensic analysis
(partial logs, unknown attacker dwell time, limited tooling)
• The host had long dwell time (weeks or months) — too many
possible persistence locations to verify exhaustively
• The host ran without EDR during any part of the incident window
• The host is a server that was potentially accessed interactively
(RDP session by attacker) — file system integrity is suspect
• The attacker had kernel-level access (rootkit indicators)
• Rapid turnaround is required and forensic validation would take
longer than re-imaging
• The host is a standard workstation (low uniqueness, golden image
available) — re-imaging is fast and certain
CLEAN IN PLACE when:
─────────────────────────────────────────────────────────────────
• Thorough forensic analysis is complete and you have high
confidence in the scope of attacker activity
• The host has complete logs covering the full incident window
• The host runs services that cannot be interrupted (production
database, critical application server) and re-imaging would
require lengthy maintenance window
• You need to preserve the production state of the host for
operational reasons and can validate clean state forensically
• The attacker's dwell time was short and well-understood
EITHER WAY:
─────────────────────────────────────────────────────────────────
→ Take a forensic image BEFORE re-imaging or cleaning
→ Verify clean state AFTER re-imaging or cleaning
→ A re-imaged host is not clean until it passes validation
(the attacker may have compromised the golden image)
Many IR teams default to "when in doubt, re-image" as though re-imaging provides a guarantee. It does not. A re-imaged host is only as clean as the image it was built from. If the golden image is outdated and the host is redeployed with unpatched vulnerabilities, the attacker can simply recompromise through the same initial access vector. Re-imaging without patching and hardening is redeployment without security improvement. Additionally, if your deployment infrastructure (MDT, SCCM, Intune) was compromised, the golden image itself may be backdoored. Verify your deployment infrastructure as part of eradication.
Persistence Class Validation
Systematic eradication validation requires checking every persistence class on every in-scope host. "Persistence class" means a category of mechanism — not a specific tool. Checking for scheduled tasks means checking every scheduled task, not just the one the attacker used.
| Persistence Class | Where to Check | Tool | What to Look For |
|---|---|---|---|
| Scheduled Tasks | Task Scheduler library; C:\Windows\System32\Tasks; C:\Windows\SysWOW64\Tasks | Autoruns, schtasks /query, Velociraptor VQL | Tasks created during incident window, tasks with encoded commands, tasks running from unusual paths, tasks with no corresponding software installation |
| Services | HKLM\SYSTEM\CurrentControlSet\Services; sc query | Autoruns, sc query, Velociraptor | Services created during incident window, services with binary paths in unusual locations, services with encoded command lines |
| Registry Run Keys | HKCU\Software\Microsoft\Windows\CurrentVersion\Run; HKLM\...\Run; RunOnce; RunServices; AppInit_DLLs; Winlogon | Autoruns, reg query, Velociraptor | New entries added during incident window, entries pointing to unusual paths or encoded commands |
| WMI Subscriptions | WMI repository: EventFilter + EventConsumer + FilterToConsumerBinding | Autoruns (WMI tab), Get-WMIObject, Velociraptor | ActiveScriptEventConsumer or CommandLineEventConsumer entries not associated with known software |
| DLL Hijacking | Application directories with writable DLLs; PATH-order hijack locations | Process Monitor (logging), manual review of application directories | DLLs placed in application directories that don't belong there; unsigned DLLs in locations that should contain signed ones |
| COM Object Hijacking | HKCU\Software\Classes\CLSID (user-level COM hijack) | Autoruns (COM tab), reg query | CLSID entries in HKCU that mirror HKLM entries and point to attacker-controlled DLLs |
| Startup Folder | %APPDATA%\Microsoft\Windows\Start Menu\Programs\Startup; %ProgramData%\...\Startup | Autoruns, manual directory listing | Files placed during incident window |
| BITS Jobs | Background Intelligent Transfer Service job queue | bitsadmin /list /allusers, Autoruns | Jobs with attacker-controlled URLs or local paths |
| LSA SSP/AP | HKLM\SYSTEM\CurrentControlSet\Control\Lsa\Security Packages and Authentication Packages | Autoruns, reg query | Unexpected DLLs added to the authentication package list (these load into LSASS) |
Active Directory Eradication Validation
Active Directory is both the highest-value target and the hardest environment to validate clean. An attacker with domain admin access for even a short period can make changes that persist long after the obvious malware is removed. AD eradication validation is its own workstream, not an afterthought.
AD Eradication Validation Checklist
═══════════════════════════════════════════════════════════════════
ACCOUNTS
□ All accounts created during incident window — documented and either
confirmed as legitimate or deleted
□ All accounts with privilege changes during incident window — reviewed
and either confirmed legitimate or rolled back
□ Admin group membership (Domain Admins, Enterprise Admins, Schema Admins,
Protected Users) — verified against pre-incident baseline
□ Service accounts — reviewed for password changes or new SPNs
(new SPN = potential Kerberoasting target the attacker set up)
GROUP POLICY
□ All GPOs created or modified during incident window — reviewed
for malicious content (startup scripts, logon scripts, scheduled tasks,
AppInit_DLLs, software installation pointing to attacker content)
□ GPO delegation — no unexpected principals with GPO edit rights
TRUSTS AND FEDERATION
□ Domain trusts — no new trusts established during incident window
□ ADFS / Entra ID federation — no new federation configuration
□ No new external connector or sync configuration
KERBEROS
□ Was KRBTGT hash accessed? (Sysmon Event 10 on DC targeting ntds.dit,
or access to KRBTGT account object)
→ If yes or unknown: rotate KRBTGT password TWICE, 10 hours apart
□ Golden ticket indicators: Kerberos tickets with anomalous PAC content,
tickets with lifetime > policy maximum
DOMAIN CONTROLLERS
□ All DCs verified: scheduled tasks, services, autoruns
□ SYSVOL — reviewed for unauthorized scripts
□ NTDS.dit — if evidence of exfiltration, assume all AD hashes are
compromised and plan full credential rotation
Sophisticated attackers approach persistence like an engineer designs fault tolerance: multiple independent mechanisms, each able to restore access if the others are removed. They assume incident response will happen and plan for it. The primary persistence is expected to be found; the secondary is designed to survive IR. Some attackers plant persistence in the BIOS firmware or the UEFI partition, knowing that re-imaging doesn't touch those. When you're validating eradication, you're not racing against the attacker's deployment — you're racing against their planning. They had more time to plan the persistence than you have to find it.
Authentication Baseline Validation
After removing persistence mechanisms, validate that authentication behavior has returned to the pre-incident baseline. The attacker's presence produces authentication anomalies; their absence should end them. If anomalies persist after eradication, either the eradication is incomplete or there is a separate, undiscovered compromise.
| Authentication signal | Baseline check | Anomaly indicates |
|---|---|---|
| Failed logon count (Event 4625) | Return to pre-incident rate | Continued credential spraying or brute force — attacker trying to regain access via a different vector |
| Explicit credential use (Event 4648) | Only from expected admin accounts during expected admin activity | Pass-the-hash or credential reuse attempt — attacker has cached credentials they're trying |
| Service ticket requests (Kerberos) | Return to baseline service access patterns per account | Kerberoasting attempts, or attacker using harvested Kerberos tickets that haven't expired |
| Remote logons to DC (Event 4624 Type 3) | Only from expected admin workstations | Continued unauthorized DC access |
| New process creation on DC (Event 4688) | Only expected administrative tools | Continued attacker execution on DC |
The Eradication Sign-Off Process
Eradication declaration should be a formal, documented act with a named individual taking responsibility for the conclusion. "The team thinks we're clean" is not a declaration — it diffuses accountability in a way that makes future reinfections impossible to learn from.
Eradication Sign-Off Requirements
═══════════════════════════════════════════════════════════════════
Required documentation before sign-off:
─────────────────────────────────────────────────────────────────
□ Scope document: every host checked, with justification for
any host excluded from scope
□ Persistence check results: per-host, per-class, with tool output
□ AD validation results: accounts, GPOs, trusts, KRBTGT status
□ Authentication baseline: confirmation that anomalies have ceased
□ Network baseline: C2 traffic has stopped (at proxy/firewall)
□ Re-imaging log (if applicable): which hosts were re-imaged, when,
from which image, patched to what level
Sign-off statement (written, timestamped, named):
─────────────────────────────────────────────────────────────────
"Based on [methodology], [person/role] declares eradication
complete as of [UTC timestamp]. Scope: [list or reference].
Known limitations: [anything that could not be verified and why].
Residual risk accepted by: [name, title]."
What triggers re-opening:
─────────────────────────────────────────────────────────────────
• New C2 traffic to known-bad infrastructure
• Authentication anomalies resuming after baseline
• New malware or tool execution from attacker-associated paths
• External notification of ongoing compromise (threat intel,
law enforcement, partner notification)
• Reappearance of indicators from the closed incident
The 2020 FireEye breach (attributed to APT29/Cozy Bear) involved a threat actor that had been operating in the environment while the IR team was actively investigating. The attackers were sophisticated enough to monitor the investigation's progress and modify their behavior to avoid detection during the active investigation phase. This demonstrates the upper bound of what a sophisticated adversary can do — they treat your IR effort as a threat to their operation and respond to it tactically. For nation-state actors, assuming the attacker doesn't know you're investigating is an assumption you cannot afford to make.
Q & A
Q: The EDR shows clean on all hosts, but the threat intel team says they still see beaconing to a known-bad domain. Which do you trust?
Trust the external signal over the internal one, and investigate the discrepancy. EDR showing clean has multiple possible explanations: the beaconing host doesn't have EDR, the EDR sensor was disabled or tampered with, or the malware is using a technique that evades the EDR (direct syscalls, process injection into a trusted process). The threat intel team seeing active beaconing is harder to fake — it means something in your environment is actively connecting to infrastructure associated with the attacker. Narrow the source: pull DNS query logs, proxy logs, and firewall flows for the past 24 hours and filter for the beaconing domain. That will identify the source IP. If the source IP maps to a host you thought was clean, re-scope that host for thorough forensics and check what process is making those connections.
Q: We re-imaged all workstations and rotated all passwords. Is that sufficient for eradication?
Not if the attacker had any access beyond workstations. Re-imaging workstations removes workstation-resident persistence. Password rotation removes the value of harvested credentials. But if the attacker reached servers, DCs, or AD, those are not cleaned by workstation re-imaging. And if they established persistence in AD (new admin accounts, WMI subscriptions on servers, GPO modifications), those persist regardless of what you do to workstations. The question to ask is: what is the furthest point the attacker reached in the environment? Eradication must cover everything at or below that watermark in the privilege hierarchy. Domain admin access → full AD validation required, every server checked, KRBTGT considered for rotation. Workstation-only access with no lateral movement → workstation re-imaging may be sufficient.
Q: How long should eradication validation take? The business wants systems back online.
The right answer is "as long as it takes to validate each host in scope, at the required depth." The practical answer depends on tooling. With Velociraptor or an EDR with fleet query capability, you can run persistence checks across hundreds of hosts in 30–60 minutes. Without fleet tooling, manual host-by-host validation at 30–45 minutes per host scales poorly against large environments. The business pressure is real but dangerous — rushing eradication is how you get a second incident two weeks later that costs far more time and money than the extra 12 hours of validation would have. The IR lead's job is to give a realistic estimate based on scope and tooling, not to promise a timeline that requires cutting corners.
Q: Should we tell the business we're in eradication, or just do the work quietly?
Communicate regularly but carefully. Business stakeholders need to know that you're in the eradication validation phase, what the timeline looks like, and what the criteria for completion are — so they can plan around the recovery timeline. What you should not communicate widely: specific details of what you found and didn't find (operational security during an active incident — the attacker may be monitoring internal communications), the specific persistence locations you've cleared vs haven't yet (same reason), and your uncertainty level about completeness (this can cause premature pressure to close). Brief the CISO and incident owner regularly on progress and estimated completion. Hold the technical details for the IR team's internal records.
Q: We found a WMI subscription on a domain controller that our eradication team missed on the first pass. How do you handle this?
Three actions simultaneously: first, remove the WMI subscription and document it (component, creator account, timestamp, payload). Second, re-evaluate the entire eradication validation for that host and any host the attacker may have accessed from the DC with the missed persistence — the missed item means your previous validation was incomplete, so the "clean" status of adjacent hosts is now uncertain. Third, update your eradication methodology to add WMI subscriptions on DCs as an explicit check item — a gap in your checklist is as important a finding as the persistence mechanism itself. The incident report should document both the finding and the methodology gap. The missed WMI subscription on the DC is a process failure, not just a technical oversight.
Q: The incident involved a nation-state actor. How does that change your eradication approach?
Substantially. Nation-state actors typically: (1) use multiple persistence mechanisms across multiple tiers of the environment; (2) have UEFI/firmware implant capability that survives re-imaging; (3) may have compromised your update/deployment infrastructure; (4) conduct counter-IR — monitoring your investigation and shifting tactics in response. For a confirmed nation-state actor, the eradication approach must include hardware-level validation (UEFI firmware integrity check on high-value hosts), full replacement of potentially compromised infrastructure (new management servers, new deployment pipelines built from verified sources), assumption that all credentials are compromised regardless of whether you found evidence of credential access, and coordination with government bodies (CISA, FBI) who may have intelligence about this actor's specific techniques. The standard eradication checklist is a floor, not a ceiling, for nation-state incidents.