Home/ATT&CK Technique/DLL Search Order Hijacking
ATT&CK Technique

DLL Search Order Hijacking

T1038 · persistence, privilege-escalation, stealth
▤ Generate a SIEM detection for T1038 ◈ Vendor-native detections for T1038 ⚠ CVEs mapped to T1038 ♛ Hunt package for T1038 ◎ Check your coverage for T1038

Windows systems use a common method to look for required DLLs to load into a program. Adversaries may take advantage of the Windows DLL search order and programs that ambiguously specify DLLs to gain privilege escalation and persistence. Adversaries may perform DLL preloading, also called binary planting attacks, by placing a malicious DLL with the same name as an ambiguously specified DLL in a location that Windows searches before the legitimate DLL.

Often this location is the current working directory of the program. Remote DLL preloading attacks occur when a program sets its current directory to a remote location such as a Web share before loading a DLL. Adversaries may use this behavior to cause the program to load a malicious DLL.

Adversaries may also directly modify the way a program loads DLLs by replacing an existing DLL or modifying a .manifest or .local redirection file, directory, or junction to cause the program to load a different DLL to maintain persistence or privilege escalation. If a search order-vulnerable program is configured to run at a higher privilege level, then the adversary-controlled DLL that is loaded will also be executed at the higher level. In this case, the technique could be used for privilege escalation from user to administrator or SYSTEM or from administrator to SYSTEM, depending on the program.

Programs that fall victim to path hijacking may appear to behave normally because malicious DLLs may be configured to also load the legitimate DLLs they were meant to replace.

Windows
How to use this page - the detection-engineering loop
Attackers have goals (tactics - “get credentials”, “move laterally”) and techniques are the concrete methods they use to reach them. This page is one method - T1038 - broken into everything you need to catch it.
The loop this page is built for (this is the job):
  1. Understand the behaviour - read the description and the Atomic Tests to see exactly what the attacker does on a host or network.
  2. Find the telemetry - what data source would reveal it (process creation, registry, network flow, auth logs). Detection Coverage shows which surfaces already have a rule and which are blind.
  3. Get or write the detection - adapt ready logic (CAR Analytics, SIEM Detections, Falco, or Sigma via Generate a SIEM detection), or author your own.
  4. Test it - run an Atomic Test in a lab and confirm your rule actually fires. A detection you have not tested is a hope, not coverage.
  5. Deploy and tune - push it, then watch for false positives and adjust.
What each panel is for:
Atomic Testssafely reproduce the technique in a lab to validate that a detection fires. Detection Coveragewhich detection surfaces have a rule for this technique; none is a blind spot to close, or simply not applicable (YARA matches files, not network behaviour). CAR / SIEM / Falcoready-made detection logic (Splunk SPL, Elastic EQL, Sentinel KQL, Falco) you adapt to your own SIEM. Mitigationsreduce exposure so the technique is harder to use at all - prevent, not just detect. Actors / Attributionwho actually uses this, so you prioritise by your own threat model. Attack Path / LOTLwhat attackers do before and after this step, and the legitimate tools they abuse to do it.
Where this fits: you usually arrive here from a CVE (“which techniques does it enable”) and leave with a tested detection deployed. The buttons above jump straight to building one, the deployable rules, the CVEs that use T1038, and a hunt package.

Detection Coverage

0/9 layers
Coverage across standard detection surfaces. Rows marked none have no rule of that type mapped. Some are real blind spots worth closing; others are simply not applicable to this technique (e.g. YARA matches malware files, not network behaviour).
Behavioral / log (Sigma) none
Analytics (MITRE CAR) none
Runtime / container (Falco) none
File / malware (YARA) none
Network (Suricata/Snort) none
Vuln scan (Nuclei) none
SIEM (Splunk ESCU) none
SIEM (Elastic) none
SIEM (Azure Sentinel) none
External lookups - second-class, for what we don’t hold ourselves