anomaly detection
Learning what normal behaviour looks like for a workload or identity and flagging what is new.
In more detail
New is not necessarily bad, so expect false positives, and relearn baselines after every deploy. An attacker who acts during the learning window becomes part of normal.
Example
An anomaly rule learns a baseline, such as how many logins an account usually makes, and alerts when today is far outside it, often measured with a z-score.Related terms
- signature-based detection — Detecting attacks by matching events against rules describing known bad behaviour.
- container drift — A running container executing something that was not in its image, such as a binary written to its writable layer after start.
- Amazon GuardDuty — AWS's managed threat detection service, which analyses CloudTrail, VPC Flow Logs, DNS logs and more to report likely attacks as findings.
- baseline — A recorded picture of what normal looks like, used as the reference for spotting changes.
- z-score — How many standard deviations a value lies from the mean: 0 is exactly average, 3 is far above normal.
- signature-based detection — Detecting attacks by matching something already known to be bad, such as a file hash or a byte pattern.
Taught in
- Testing Resilience: The Experiment That Would Have Been The Outage — AWS Certified DevOps Engineer – Professional (DOP-C02) Exam Prep
- Finding The Problem: The Composite Alarm That Stayed Quiet — AWS Certified DevOps Engineer – Professional (DOP-C02) Exam Prep
- Cost Governance: Tags, Budgets And Reports That Speak Up — AWS Certified Solutions Architect – Associate (SAA-C03) Exam Prep
- Runtime Anomaly Detection with Falco & eBPF — Cloud Security & DevSecOps Engineering
- Threat Detection with GuardDuty & Anomaly Analytics — Cloud Security & DevSecOps Engineering
- SIEM Architecture: Ingestion Pipelines & Indexing — Security Operations & Threat Hunting (SOC)
- Event Correlation Rules, Thresholds & Anomaly Pipelines — Security Operations & Threat Hunting (SOC)