Group related security alerts into attack stories
An agent connects scattered alerts across hosts and users into likely attack chains mapped to MITRE ATT&CK for analyst review.
The use case
An attacker's steps often show up as unrelated low-severity alerts on different hosts. Elastic's Attack Discovery uses language models to analyze open alerts and group related ones into a narrative, listing the affected accounts and machines and mapping the activity to MITRE ATT&CK. Analysts can spot a campaign that would otherwise sit across a dozen tickets. The analyst still validates each story against raw events, decides whether it is a real incident, and leads the response. Teams often run it at the start of each shift on the last day of alerts, so analysts begin with a short list of possible campaigns.
Example workflow
- Run Attack Discovery on recent open alerts
- Read each grouped attack story
- Check the underlying events for each step
- Open an incident for confirmed stories
- Feed results back into detection tuning
Tools
- Elastic Security
Risks and guardrails
- Grouping can link unrelated alerts; verify each step
- Model costs grow with alert volume; scope the time window
- Do not take response actions on an unverified story
Sources
Written by Agents Lookup. Credits and official pages for the tools this use case names:
- Elastic docs: Attack Discovery: www.elastic.co/docs/solutions/security/ai/attack-discovery