Security alert triage before an analyst picks it up
An agent investigates each SOC alert, gathers evidence, and gives a verdict with reasoning so analysts start from a summary.
The use case
Security operations teams drown in alerts, many of them benign. Triage agents such as Dropzone AI, Google Security Operations' triage and investigation agent and CrowdStrike's Charlotte AI investigate alerts as they arrive: they look up related logs, user and host details and threat intel, then write up a verdict with the evidence. Analysts start from that write-up instead of a raw alert. The analyst still checks the verdict on anything important, decides on containment such as isolating a host or disabling an account, and reviews a sample of benign verdicts to catch misses.
Example workflow
- Connect the agent to SIEM, EDR and identity logs
- Agent investigates new alerts and writes a verdict
- Analyst reviews malicious and uncertain verdicts first
- Analyst decides on containment actions
- Sample benign verdicts weekly for quality
Tools
- Dropzone AI
- Google Security Operations
- CrowdStrike Charlotte AI
Risks and guardrails
- A wrong benign verdict can hide a real attack; audit samples
- Containment actions should need analyst approval
- Attackers may plant text in logs to mislead the agent
Sources
Written by Agents Lookup. Credits and official pages for the tools this use case names:
- Dropzone AI: www.dropzone.ai
- Google Cloud docs: SecOps triage and investigation agent: docs.cloud.google.com/chronicle/docs/secops/triage-investigation-agent
- CrowdStrike: Charlotte AI: www.crowdstrike.com/en-us/platform/charlotte-ai