Week 13 · slides

Week 13

← all weeks · worksheet · readme · harden

marp: true theme: default paginate: true header: "Software Security · Week 13"

Week 13

Cloud & Container Security

Software Security · Nutthakorn Chalaemwongwan

<!-- Hook: most cloud breaches aren't clever zero-days — they're a public S3 bucket or a *:* IAM policy. Today we hunt and fix those misconfigs. ~2 min. -->

Today

  • Shared-responsibility model
  • IAM & least privilege
  • Secrets management
  • Container/image hardening
  • 🎮 Game: Misconfig Hunt

<!-- Roadmap, 1 min. Theme: the cloud is secure; your CONFIGURATION usually isn't. Lab = find + fix misconfigurations, each one a flag. -->

Recap & framing

  • Supply chain → what you build with
  • Today → where you run it
  • OWASP A02:2025 Security Misconfiguration (now #2)

<!-- 1-min bridge. Misconfiguration jumped to A02 in 2025 — it's not a footnote, it's the second most critical web risk. ~2 min. -->

Shared responsibility

  • Cloud secures of the cloud; you secure in the cloud
  • Misconfig — not provider bugs — causes most breaches
  • Defaults are rarely safe

<!-- The mental model students most misunderstand: AWS secures the hardware/hypervisor; YOU secure your buckets, IAM, security groups. Almost every headline cloud breach is the customer's config, not the provider. ~4 min. -->

IAM & least privilege

{ "Effect":"Allow", "Action":"*", "Resource":"*" }   // 🚩
  • Over-broad policies = blast radius
  • Grant only the actions/resources needed
  • CWE-732 (bad permissions), CWE-16 (config)

<!-- The worked example — this *:* policy is the Misconfig Hunt round 1 and the quiz. If a credential with this leaks, the attacker owns everything. Ask the class to rewrite it as read-only on one bucket. Least privilege = smallest possible blast radius. ~6 min. -->

Secrets management

  • Secrets in env vars / Dockerfile / git = leaked
  • Use a secrets manager / vault; rotate
  • Scan history (Gitleaks) — recall Week 2

<!-- Connect to W2 (Gitleaks) and W12. A secret baked into an image layer is in every copy of that image forever — docker history reveals it. Fix = inject at runtime from a vault + rotate. ~4 min. -->

Storage & network exposure

  • Public buckets, open ports, default creds
  • Encrypt at rest + in transit
  • Private by default; explicit allow

<!-- The classic trio behind most breaches. 0.0.0.0/0 on a DB port = the internet can reach your database. Default creds = free entry. Principle: deny by default, open deliberately (echoes W6 access control). ~4 min. -->

Container image hardening

trivy config /src       # IaC/Dockerfile misconfig
trivy image myapp:lab   # image CVEs
  • Minimal/distroless base, drop root, pin versions
  • Re-scan to prove reduced findings

<!-- Hands-on — this is the lab's insecure→hardened Dockerfile. trivy config finds the misconfig (root user, latest tag); trivy image finds CVEs in the base. Drop root + distroless = container escape gives far less. Re-scan to prove improvement. ~5 min. -->

Kubernetes basics (awareness)

  • Pod security, network policies, RBAC
  • Don't run privileged; limit service-account tokens

<!-- Awareness only (K8s is its own course). Key takeaways: a privileged pod ≈ host root; mounted service-account tokens are a lateral-movement prize. Keep it brief. ~3 min. -->

🔍 Game — Misconfig Hunt (CloudGoat-style)

Each misconfiguration found + fixed = a flag:

  1. Over-permissive IAM (*:*) → least privilege
  2. Exposed bucket → lock down
  3. Env-var secrets → secrets manager
  4. Vulnerable Dockerfile → harden + re-scan

<!-- Explain before lab. Find AND fix = the flag (defending is the point). Q6 of the quiz asks for one misconfig they fixed + the principle it violated. ~3 min. -->

Deliverable

📋 Worksheet 13 — labs/week13-cloud-container/worksheet.md (Part 3) · kickoff: bash scan.sh (trivy config over Dockerfiles + IAM JSON)

  • Before/after IAM policy + Dockerfile
  • Trivy reports showing reduced risk
  • Note on secrets remediation
  • + Audit the AI / EiPE / Prompt Problem (see worksheet)

<!-- Before/after artifacts + the trivy delta prove the fix. AI-resilient tasks count. -->

Key takeaways

  • Misconfiguration > zero-days as a breach cause
  • Least privilege, private-by-default, no secrets in code
  • Scan IaC and images in CI

<!-- Recap. Cold-call: "what does a *:* IAM policy cost you if the key leaks?" (everything — full blast radius). "Scan in CI" sets up W15. ~2 min. -->

Questions?

Next week: AI / LLM application security

<!-- Cliffhanger: "Next week — the newest attack surface: make an AI assistant ignore its rules and leak secrets, with nothing but text." -->