Windows Local LLM Framework Execution
SPLUNK detection rule by Rod Soto, Nasreddine Bencherchali, Splunk · Apache-2.0
The following analytic detects execution of unauthorized local LLM frameworks (Ollama, LM Studio, GPT4All, Jan, llama.cpp, KoboldCPP, Oobabooga, NutStudio) and Python-based AI/ML libraries (HuggingFace Transformers, LangChain) on Windows endpoints by leveraging process creation events.
It identifies cases where known LLM framework executables are launched or command-line arguments reference AI/ML libraries.
This activity is significant as it may indicate shadow AI deployments, unauthorized model inference operations, or potential data exfiltration through local AI systems.
If confirmed malicious, this could lead to unauthorized access to sensitive data, intellectual property theft, or circumvention of organizational AI governance policies.
name: Windows Local LLM Framework Execution
id: a3f8e2c9-7d4b-4e1f-9c6a-2b5d8f3e1a7c
version: 3
creation_date: '2025-11-24'
modification_date: '2026-07-15'
author: Rod Soto, Nasreddine Bencherchali, Splunk
status: production
type: Hunting
description: |
The following analytic detects execution of unauthorized local LLM frameworks (Ollama, LM Studio, GPT4All, Jan, llama.cpp, KoboldCPP, Oobabooga, NutStudio) and Python-based AI/ML libraries (HuggingFace Transformers, LangChain) on Windows endpoints by leveraging process creation events.
It identifies cases where known LLM framework executables are launched or command-line arguments reference AI/ML libraries.
This activity is significant as it may indicate shadow AI deployments, unauthorized model inference operations, or potential data exfiltration through local AI systems.
If confirmed malicious, this could lead to unauthorized access to sensitive data, intellectual property theft, or circumvention of organizational AI governance policies.
data_source:
- Sysmon EventID 1
- Windows Event Log Security 4688
- CrowdStrike ProcessRollup2
search: |
| tstats `security_content_summariesonly` count
min(_time) as firstTime
max(_time) as lastTime
from datamodel=Endpoint.Processes
where
(
Processes.process_name IN (
"gpt4all.exe",
"jan.exe",
"kobold.exe",
"koboldcpp.exe",
"llama-run.exe",
"llama.cpp.exe",
"lmstudio.exe",
"nutstudio.exe",
"ollama.exe",
"oobabooga.exe",
"text-generation-webui.exe"
)
OR
Processes.original_file_name IN (
"gpt4all.exe",
"jan.exe",
"koboldcpp.exe",
"llama-run.exe",
"lmstudio.exe",
"nutstudio.exe",
"ollama.exe"
)
OR
Processes.process IN (
"*\\gpt4all\\*",
"*\\jan\\*",
"*\\koboldcpp\\*",
"*\\llama.cpp\\*",
"*\\lmstudio\\*",
"*\\nutstudio\\*",
"*\\ollama\\*",
"*\\oobabooga\\*",
"*huggingface*",
"*langchain*",
"*llama-run*",
"*transformers*"
)
OR
Processes.parent_process_name IN (
"gpt4all.exe",
"jan.exe",
"kobold.exe",
"koboldcpp.exe",
"llama-run.exe",
"llama.cpp.exe",
"lmstudio.exe",
"nutstudio.exe",
"ollama.exe",
"oobabooga.exe",
"text-generation-webui.exe"
)
)
NOT Processes.process_name IN (
"360se.exe", "Acrobat.exe", "arc.exe", "atlas.exe", "avastbrowser.exe",
"avgbrowser.exe", "blpbrowser.exe", "brave.exe", "braveupdate.exe",
"chrome.exe", "chromium.exe", "coccoc.exe", "comet.exe", "conhost.exe",
"dragon.exe", "duckduckgo.exe", "epic.exe", "EXCEL.EXE", "firefox.exe",
"FoxitPDFReader.exe", "ghost.exe", "iexplore.exe", "iridium.exe", "iron.exe",
"maxthon.exe", "microsoftedge.exe", "microsoftedgeupdate.exe", "msedge.exe",
"msedgewebview2.exe", "mspaint.exe", "NewformaOCR.exe", "notepad.exe",
"opera_autoupdate.exe", "opera.exe", "OUTLOOK.EXE", "perplexity.exe",
"Photos.exe", "POWERPNT.EXE", "rave.exe", "Revu.exe", "safari.exe",
"shift.exe", "sidekick.exe", "Stapler.exe", "thorium.exe", "torch.exe",
"vivaldi.exe", "wavebrowser.exe", "whale.exe", "WINWORD.EXE", "zen.exe"
)
by Processes.action Processes.dest Processes.original_file_name Processes.parent_process
Processes.parent_process_exec Processes.parent_process_guid Processes.parent_process_id
Processes.parent_process_name Processes.parent_process_path Processes.process
Processes.process_exec Processes.process_guid Processes.process_hash Processes.process_id
Processes.process_integrity_level Processes.process_name Processes.process_path Processes.user
Processes.user_id Processes.vendor_product
| `drop_dm_object_name(Processes)`
| eval Framework=case(
match(process_name, "(?i)ollama") OR match(process, "(?i)ollama"), "Ollama",
match(process_name, "(?i)lmstudio") OR match(process, "(?i)lmstudio") OR match(process, "(?i)lm-studio"), "LM Studio",
match(process_name, "(?i)gpt4all") OR match(process, "(?i)gpt4all"), "GPT4All",
match(process_name, "(?i)kobold") OR match(process, "(?i)kobold"), "KoboldCPP",
match(process_name, "(?i)jan") OR match(process, "(?i)jan"), "Jan AI",
match(process_name, "(?i)nutstudio") OR match(process, "(?i)nutstudio"), "NutStudio",
match(process_name, "(?i)llama") OR match(process, "(?i)llama"), "llama.cpp",
match(process_name, "(?i)oobabooga") OR match(process, "(?i)oobabooga") OR match(process, "(?i)text-generation-webui"), "Oobabooga",
match(process, "(?i)transformers") OR match(process, "(?i)huggingface"), "HuggingFace/Transformers",
match(process, "(?i)langchain"), "LangChain",
1=1, "Other"
)
| `security_content_ctime(firstTime)`
| `security_content_ctime(lastTime)`
| table action dest Framework original_file_name parent_process parent_process_exec
parent_process_guid parent_process_id parent_process_name parent_process_path
process process_exec process_guid process_hash process_id process_integrity_level
process_name process_path user user_id vendor_product
| `windows_local_llm_framework_execution_filter`
how_to_implement: |
The detection is based on data that originates from Endpoint Detection
and Response (EDR) agents. These agents are designed to provide security-related
telemetry from the endpoints where the agent is installed. To implement this search,
you must ingest logs that contain the process GUID, process name, and parent process.
Additionally, you must ingest complete command-line executions. These logs must
be processed using the appropriate Splunk Technology Add-ons that are specific to
the EDR product. The logs must also be mapped to the `Processes` node of the `Endpoint`
data model. Use the Splunk Common Information Model (CIM) to normalize the field
names and speed up the data modeling process.
known_false_positives: Legitimate development, data science, and AI/ML workflows where authorized developers, researchers, or engineers intentionally execute local LLM frameworks (Ollama, LM Studio, GPT4All, Jan, NutStudio) for model experimentation, fine-tuning, or prototyping. Python developers using HuggingFace Transformers or LangChain for legitimate AI/ML projects. Approved sandbox and lab environments where framework testing is authorized. Open-source contributors and hobbyists running frameworks for educational purposes. Third-party applications that bundle or invoke LLM frameworks as dependencies (e.g., IDE plugins, analytics tools, chatbot integrations). System administrators deploying frameworks as part of containerized services or orchestrated ML workloads. Process name keyword overlap with unrelated utilities (e.g., "llama-backup", "janimation"). Recommended tuning — baseline approved frameworks and users by role/department, exclude sanctioned dev/lab systems via the filter macro, correlate with user identity and peer group anomalies before escalating to incident response.
references:
- https://splunkbase.splunk.com/app/8024
- https://www.ibm.com/think/topics/shadow-ai
- https://www.splunk.com/en_us/blog/artificial-intelligence/splunk-technology-add-on-for-ollama.html
- https://blogs.cisco.com/security/detecting-exposed-llm-servers-shodan-case-study-on-ollama
- https://docs.microsoft.com/en-us/sysinternals/downloads/sysmon
analytic_story:
- Suspicious Local LLM Frameworks
asset_type: Endpoint
mitre_attack_id:
- T1543
product:
- Splunk Enterprise
- Splunk Enterprise Security
- Splunk Cloud
category: endpoint
security_domain: endpoint
tests:
- name: True Positive Test - Sysmon
attack_data:
- data: https://media.githubusercontent.com/media/splunk/attack_data/master/datasets/suspicious_behaviour/local_llms/sysmon_local_llms.log
source: XmlWinEventLog:Microsoft-Windows-Sysmon/Operational
sourcetype: XmlWinEventLog
test_type: unit
CIRCL