Branch was auto-updated.

2026-06-08 17:32:49 +00:00 · 2022-07-19 10:46:05 -07:00
parent c12492e1ee a0842978fa
commit a9ec0428ff
4 changed files with 80 additions and 0 deletions
@@ -0,0 +1,58 @@
+name: Detect Risky SPL using Pretrained ML Model
+id: b4aefb5f-1037-410d-a149-1e091288ba33
+version: 1
+date: '2022-06-16'
+author: Abhinav Mishra, Kumar Sharad, Namratha Sreekanta and Xiao Lin, Splunk
+type: Anomaly
+datamodel:
+- Splunk_Audit
+description: 'The following analytic uses a pretrained machine learning text classifier to detect potentially risky commands. The model is trained independently and then the model file is packaged within ESCU for usage. A command is deemed risky based on the presence of certain trigger keywords, along with the context and the role of the user (please see references). The model uses custom features to predict whether a SPL is risky using text classification. The model takes as input the command text, user and search type and outputs a risk score between [0,1]. A high score indicates higher likelihood of a command being risky. This model is on-prem only.'
+search: "| tstats `security_content_summariesonly` count min(_time) as firstTime max(_time) as lastTime from datamodel=Splunk_Audit.Search_Activity where Search_Activity.search_type=adhoc Search_Activity.user!=splunk-system-user by Search_Activity.search Search_Activity.user Search_Activity.search_type | eval spl_text = 'Search_Activity.search'. \" \" .'Search_Activity.user'. \" \" .'Search_Activity.search_type'| dedup spl_text 
+| apply risky_spl_pre_trained_model 
+| where risk_score > 0.5 
+| `drop_dm_object_name(Search_Activity)` 
+| table search, user, search_type, risk_score 
+| `detect_risky_spl_using_pretrained_ml_model_filter`" 
+how_to_implement: This detection depends on the MLTK app which can be found here - https://splunkbase.splunk.com/app/2890/ and the Splunk Audit datamodel which can be found here - https://splunkbase.splunk.com/app/1621/. Additionally, you need to be ingesting logs which include Search_Activity.search, Search_Activity.user, Search_Activity.search_type from your endpoints. The risk score threshold should be adjusted based on the environment. The detection uses a custom MLTK model hence we need a few more steps for deployment, as outlined here - https://gist.github.com/ksharad-splunk/be2a62227966049047f5e5c4f2adcabb.
+known_false_positives: False positives may be present if suspicious behavior is observed, as determined by frequent usage of risky keywords.
+references:
+- https://docs.splunk.com/Documentation/Splunk/latest/Security/SPLsafeguards#Commands_that_trigger_the_warning
+tags:
+  analytic_story:
+  - Splunk Vulnerabilities
+  asset_type: Web Server
+  cis20:
+  - CIS 3
+  - CIS 6
+  confidence: 40
+  cve:
+  - CVE-2022-32154
+  context:
+  - Source:Endpoint
+  dataset:
+  - https://github.com/splunk/attack_data/raw/master/datasets/attack_techniques/T1203/search_activity.txt
+  impact: 50
+  kill_chain_phases:
+  - Actions on Objectives
+  message: A potentially risky Splunk command has been run by $user$, kindly review.
+  mitre_attack_id:
+  - T1059
+  nist:
+  - DE.AE
+  observable:
+  - name: user
+    type: User
+    role:
+    - Victim
+  product:
+  - Splunk Enterprise
+  - Splunk Enterprise Security
+  - Splunk Cloud
+  required_fields:
+  - _time
+  - Search_Activity.search 
+  - Search_Activity.user 
+  - Search_Activity.search_type
+  risk_score: 20
+  security_domain: audit
+  
@@ -0,0 +1,2 @@
+algo,model,options
+DetectRiskySPL,"{""__mlspl_type"": [""algos.DetectRiskySPL"", ""DetectRiskySPL""], ""dict"": {""classes"": null, ""target_variable"": [""risk_score""], ""feature_variables"": [""spl_text""], ""columns"": [""spl_text""], ""estimator"": {""__mlspl_type"": [""sklearn.pipeline"", ""Pipeline""], ""dict"": {""steps"": [[""features"", {""__mlspl_type"": [""sklearn.feature_extraction.text"", ""CountVectorizer""], ""dict"": {""input"": ""content"", ""encoding"": ""utf-8"", ""decode_error"": ""strict"", ""strip_accents"": null, ""preprocessor"": null, ""tokenizer"": null, ""analyzer"": ""word"", ""lowercase"": true, ""token_pattern"": "" collect | delete | fit | outputcsv | outputlookup |adhoc| sendalert | sendemail |splunk\\-system\\-user| tscollect | run | script | runshellscript "", ""stop_words"": null, ""max_df"": 1.0, ""min_df"": 1, ""max_features"": null, ""ngram_range"": [1, 1], ""vocabulary"": null, ""binary"": false, ""dtype"": {""__mlspl_type"": [""builtins"", ""type""], ""type"": [""numpy"", ""int64""]}, ""fixed_vocabulary_"": false, ""_stop_words_id"": 94300723879360, ""stop_words_"": {""__mlspl_type"": [""builtins"", ""set""], ""set"": []}, ""vocabulary_"": {""splunk-system-user"": 12, "" delete "": 1, ""adhoc"": 11, "" outputlookup "": 4, "" script "": 7, "" run "": 5, "" collect "": 0, "" sendemail "": 9, "" sendalert "": 8, "" outputcsv "": 3, "" fit "": 2, "" runshellscript "": 6, "" tscollect "": 10}}}], [""predictor"", {""__mlspl_type"": [""sklearn.linear_model._logistic"", ""LogisticRegression""], ""dict"": {""penalty"": ""l2"", ""dual"": false, ""tol"": 0.0001, ""C"": 1.0, ""fit_intercept"": true, ""intercept_scaling"": 1, ""class_weight"": {""0"": 1, ""1"": 10}, ""random_state"": null, ""solver"": ""liblinear"", ""max_iter"": 100, ""multi_class"": ""auto"", ""verbose"": 0, ""warm_start"": false, ""n_jobs"": null, ""l1_ratio"": null, ""n_features_in_"": 13, ""classes_"": {""__mlspl_type"": [""numpy"", ""ndarray""], ""npy"": ""k05VTVBZAQB2AHsnZGVzY3InOiAnPGk4JywgJ2ZvcnRyYW5fb3JkZXInOiBGYWxzZSwgJ3NoYXBlJzogKDIsKSwgfSAgICAgICAgICAgICAgICAgICAgICAgICAgICAgICAgICAgICAgICAgICAgICAgICAgICAgICAgICAgIAoAAAAAAAAAAAEAAAAAAAAA""}, ""coef_"": {""__mlspl_type"": [""numpy"", ""ndarray""], ""npy"": ""k05VTVBZAQB2AHsnZGVzY3InOiAnPGY4JywgJ2ZvcnRyYW5fb3JkZXInOiBGYWxzZSwgJ3NoYXBlJzogKDEsIDEzKSwgfSAgICAgICAgICAgICAgICAgICAgICAgICAgICAgICAgICAgICAgICAgICAgICAgICAgICAgICAgIAqulbT8VG8TQJU6VfC9QuY/kCCmapJVFUDQl14TS2ApPw5vYc32jBxAxVuQ3Sv35D8Y+azG/kDmP9vpUE0rTwlALsMVcoUGE0ASjjFaKyMaQA2zZ/yMQRZAQLZHc97OHEAfrzTDBGwSwA==""}, ""intercept_"": {""__mlspl_type"": [""numpy"", ""ndarray""], ""npy"": ""k05VTVBZAQB2AHsnZGVzY3InOiAnPGY4JywgJ2ZvcnRyYW5fb3JkZXInOiBGYWxzZSwgJ3NoYXBlJzogKDEsKSwgfSAgICAgICAgICAgICAgICAgICAgICAgICAgICAgICAgICAgICAgICAgICAgICAgICAgICAgICAgICAgIAqBnhyKtBckwA==""}, ""n_iter_"": {""__mlspl_type"": [""numpy"", ""ndarray""], ""npy"": ""k05VTVBZAQB2AHsnZGVzY3InOiAnPGk0JywgJ2ZvcnRyYW5fb3JkZXInOiBGYWxzZSwgJ3NoYXBlJzogKDEsKSwgfSAgICAgICAgICAgICAgICAgICAgICAgICAgICAgICAgICAgICAgICAgICAgICAgICAgICAgICAgICAgIAoLAAAA""}}}]], ""memory"": null, ""verbose"": false}}}}","{""args"": [""risk_score"", ""spl_text""], ""target_variable"": [""risk_score""], ""feature_variables"": [""spl_text""], ""model_name"": ""risky_spl_pre_trained_model"", ""algo_name"": ""LogisticRegression"", ""mlspl_limits"": {""handle_new_cat"": ""default"", ""max_distinct_cat_values"": ""100"", ""max_distinct_cat_values_for_classifiers"": ""100"", ""max_distinct_cat_values_for_scoring"": ""100"", ""max_fit_time"": ""600"", ""max_inputs"": ""100000"", ""max_memory_usage_mb"": ""1024"", ""max_model_size_mb"": ""15"", ""max_score_time"": ""600"", ""streaming_apply"": ""false"", ""use_sampling"": ""true""}, ""kfold_cv"": null}"
@@ -0,0 +1,6 @@
+description: Detect Risky SPL using Pretrained ML Model
+filename: __mlspl_risky_spl_pre_trained_model.mlmodel
+name: __mlspl_risky_spl_pre_trained_model
+case_sensitive_match: 'false'
+min_matches: 1
+default_match: 'false'
@@ -0,0 +1,14 @@
+name: Detect Risky SPL using Pretrained ML Model Unit Test
+tests:
+- name: Detect Risky SPL using Pretrained ML Model
+  file: application/detect_risky_spl_using_pretrained_ml_model.yml
+  pass_condition: '| stats count | where count > 0'
+  earliest_time: -10y
+  latest_time: now
+  attack_data: 
+  - file_name: search_activity.txt
+    data: https://github.com/splunk/attack_data/raw/master/datasets/attack_techniques/T1203/search_activity.txt
+    source: audittrail
+    sourcetype: audittrail
+    update_timestamp: true
+    custom_index: _audit