Splunk Retires MLTK Detections for Risky SPL Command Abuse After Vulnerability Patches
Patch & Detection Intelligence · 5 sources · Updated 3mo ago · First seen May 14, 2026

Splunk has removed several Enterprise Security detections that were built to identify abuse of risky Search Processing Language (SPL) commands, including "Splunk Command and Scripting Interpreter Risky SPL MLTK", "Detect Risky SPL using Pretrained ML Model", and "Splunk risky Command Abuse disclosed february 2023". The analytics monitored Splunk_Audit.Search_Activity data for ad hoc searches using commands such as collect, delete, fit, outputcsv, outputlookup, run, script, sendalert, sendemail, and tscollect, which could be abused to bypass warnings, escalate privileges, delete data, exfiltrate information, or support arbitrary code execution and persistence. Splunk said the detections were disabled by default and mapped to ATT&CK techniques including T1059, T1202, and T1548.
The retired content relied on the Machine Learning Toolkit (MLTK) and, in some cases, pretrained models or per-user baselines to flag anomalous long-running risky searches or suspicious command text. Splunk originally promoted the approach as a way to complement rule-based detections for exploitation tied to CVE-2022-32154 and later risky-command vulnerabilities, but the company has since withdrawn the analytics from its Threat Research library because the affected flaws have been patched in current Splunk releases. Splunk also noted operational drawbacks, including false positives from legitimate long-running searches and added compute requirements for model training in larger environments.


