Skip to main content

Detection tuning

How an AI SOC tunes detections from triage verdicts: propose exclusions from closed false positives, backtest them, and ship the change for review.

Use case
Updated

The short answer

Detection tuning cuts false positives without losing real detections. In Cotool, agents investigate every alert and each verdict feeds back into the detection that fired it, so a tuning agent proposes changes from false positives closed in triage.[1]

The problem

Rule-based detections need constant tuning as environments and threats change.[2]

In Splunk, scheduled detections are saved searches, so a native rule is tuned by editing its saved search.[3][8]

How an AI SOC handles it

  1. Collect verdicts

    Agents investigate each alert and close false positives with their evidence.[1][4]

  2. Propose a change

    A tuning agent reads the closed false positives and proposes a code change, such as excluding OAuth apps on the security team's approved register.[1]

  3. Backtest it

    The change replays over 30 days of history to show how many alerts drop and that the true positives still fire.[1][5]

  4. Review and ship

    The change opens as a pull request in your repository, or is sent for review or applied from the detection page.[1][6]

  5. Keep library updates

    Customized library detections still pick up new library versions through auto-tuning, and new tunings keep changes approved earlier.[7]

  6. Tune native rules

    For detections that stay in Splunk, agents list saved searches and can create new ones with scheduling and alerting settings.[3][8]

Example

A proposed tuning for an OAuth grant detectionpython
alerts = grants.where(scope__in=SENSITIVE_SCOPES)
# Security's register of approved OAuth apps
trusted = notion.query("Approved apps")["Client ID"]
alerts = alerts.exclude(client_id__in=trusted)

The tuning a Cotool tuning agent proposed on the Detect page, built from false positives closed in triage: sensitive-scope grants still alert unless the app is on the approved register.[1]

What people approve

  • Tunings go through review, as a pull request approval or Send for review, before they apply.[1]

How Cotool handles it

  • Cotool's detection library covers common log sources, maps to MITRE ATT&CK, and is tuned to your environment by agents.[1]

Integrations used

Sources

  1. Detection engineering (Cotool Detect) · accessed November 16, 2026
  2. Pattern-based detection with agents (Cotool docs) · accessed November 16, 2026
  3. Splunk integration (Cotool docs) · accessed November 16, 2026
  4. Alerts (Cotool docs) · accessed November 16, 2026
  5. A SIEM built for AI-native security operations (Cotool SIEM) · accessed November 16, 2026
  6. GitHub integration (Cotool docs) · accessed November 16, 2026
  7. Changelog (Cotool docs) · accessed November 16, 2026
  8. Search endpoint descriptions (Splunk REST API Reference) · accessed November 16, 2026

Related use cases

All use cases

Related reading

Attackers are scaling with tokens

Cotool helps defenders operate at machine speed. See how security teams are scaling Detection & Response.

Book a demo