Cloud Security Lab

Problem

Detection rules are only trustworthy if you have seen them fire against a real attack. This lab builds both sides: an offensive kill chain and the defensive controls that should catch it, so detection efficacy is demonstrated rather than assumed.

Attack

• The full MITRE ATT&CK kill chain is scripted under attack/ so it runs identically every time and always under the leaked credential rather than an admin identity: initial access, permission enumeration, then privilege escalation from 1,039 to 15,319 permissions via policy attachment.
• Then S3 exfiltration of staged PII and lateral movement via STS role assumption, with Pacu and ScoutSuite available for deeper manual exploration.

Detect & Respond

• CloudTrail and VPC Flow Logs feed an OpenSearch SIEM with a kill-chain correlation dashboard.
• GuardDuty findings on IAM threats trigger an EventBridge rule that fires a Lambda to automatically disable compromised access keys.
• On Kubernetes, Falco runs as a DaemonSet catching runtime attacks: shell spawning, sensitive file reads, unauthorized binary execution, and container escape via host mount.
• OPA Gatekeeper blocks privileged containers, host namespace access, and root execution across all non-system namespaces.

Detection as Code

• The Gatekeeper admission policies are unit-tested with gator against known-good and known-bad pods, so a broken policy fails CI before it ever reaches a cluster.
• CI gates selectively against a codebase that is vulnerable on purpose: the defensive modules are held to a Checkov baseline that blocks new misconfigurations, the intentionally-vulnerable surface is scanned informationally, and secret scanning blocks everywhere.
• The exercised techniques are captured as an importable MITRE ATT&CK Navigator layer.

Outcome

A closed loop from exploitation to automated containment, with every control demonstrated against a live, reproducible attack rather than described in the abstract.

62 Terraform resources across 7 modules, deployable and destroyable on demand.