Skip to content

Data Protection and Workload Security

Last reviewed: August 2026

Cloud security broadly falls into three areas.

  • Security in transit — protecting data as it moves across the network
  • Security at rest — protecting data stored in storage/databases
  • Workload security — protecting running servers/containers/applications

Protects against eavesdropping or tampering as data moves across the network.

Segment Method Notes
User ↔ service TLS/HTTPS TLS terminated at the CDN or load balancer. Free certificates available (ACM, Let’s Encrypt, etc.)
Service ↔ service mTLS, VPC-internal communication A service mesh (Istio, App Mesh) can automate mTLS
Region ↔ region Vendor backbone encryption AWS/Azure/Google Cloud all automatically encrypt inter-region traffic
On-premises ↔ cloud VPN (IPsec) / dedicated line Direct Connect, ExpressRoute, Cloud Interconnect

Communication between managed services is encrypted with TLS by default across all vendors.

Encrypts stored data so it cannot be read even if it is physically stolen. All major CSPs provide default encryption, and the level of key management chosen affects the balance between security and operational complexity:

  • Vendor-managed keys — simplest. The vendor generates/rotates/manages the keys. Suitable for most workloads.
  • Customer-managed keys (CMK) — you control the rotation cadence and access policies directly. Meets regulatory requirements.
  • Your own keys (BYOK/EKM/HYOK) — keys are managed in an on-premises HSM. Suitable for the strictest regulatory requirements.

The algorithms used today for HTTPS, VPN, and data encryption — RSA, ECDH, and the like — could be broken once quantum computers become large enough. The problem isn’t “after” quantum computers arrive — it’s “now.” Attackers can already collect encrypted traffic today and decrypt it later once a sufficiently powerful quantum computer exists — a Harvest Now, Decrypt Later attack.

In response, NIST has finalized new cryptographic standards, and major cloud vendors have begun their transition.

Standard Purpose Algorithm Status
FIPS 203 (ML-KEM) Key exchange Kyber-based lattice cryptography Finalized August 2024
FIPS 204 (ML-DSA) Digital signatures Dilithium-based lattice cryptography Finalized August 2024
FIPS 205 (SLH-DSA) Digital signatures (stateless) SPHINCS+-based hash signatures Finalized August 2024
HQC Backup KEM Code-based cryptography Selected March 2025
Vendor Status Reference
AWS KMS supports ML-KEM hybrid TLS. PQ hybrid key exchange being applied to inter-service communication for S3, ACM, and others AWS Post-Quantum Cryptography
Azure Microsoft Quantum Safe Program. ML-KEM/ML-DSA implemented in the SymCrypt library. TLS 1.3 hybrid key exchange supported Microsoft Quantum Safe
Google Cloud PQC digital signatures (ML-DSA) in preview in Cloud KMS. ML-KEM hybrid deployment completed in Chrome/BoringSSL Google Cloud PQC
OCI Has announced a PQC algorithm roadmap for OCI Vault and other services. Confirm hybrid TLS mode for Oracle Database and similar items against the official roadmap/release notes (link to be updated once specific product documentation is published) Oracle Security (general hub; check official documentation for a dedicated PQC page)
  1. Inventory — identify the encryption algorithms, certificates, and key sizes in use (a Crypto Agility Inventory)
  2. Hybrid mode — use existing algorithms alongside PQC algorithms simultaneously to maintain compatibility during the transition
  3. Prioritization — transition long-retention data (10+ year lifespan) and signing infrastructure first
  4. Testing — since PQC algorithms have larger key/signature sizes, measure the network overhead and handshake latency impact

While traditional encryption protects data “at rest” and “in transit,” confidential computing protects data “in use.” Data is processed inside a hardware-based Trusted Execution Environment (TEE), so even the cloud vendor’s administrators cannot access the data being processed.

Vendor Product GPU confidential computing Reference
AWS Nitro Enclaves — (the Nitro architecture itself provides hypervisor-level isolation) AWS Nitro
Azure Confidential VMs (AMD SEV-SNP, Intel TDX) NCC H100 v5 — NVIDIA H100 confidential GPU Azure Confidential Computing
Google Cloud Confidential VMs (AMD SEV, Intel TDX) A3 Confidential VM — H100 confidential GPU GCP Confidential Computing
OCI Confidential Computing (AMD SEV) OCI Compute

Key use cases:

  • Protecting both model IP and input data simultaneously in AI/ML inference (confidential GPUs)
  • Multi-party data analysis — joint computation without exposing the underlying raw data
  • Workloads with strict data sovereignty requirements (sovereign cloud combined with confidential computing)

Protects running infrastructure and applications.

Protecting data in transit requires network-level access control.

Vendor Product Notes
AWS AWS WAF Attaches to ALB/CloudFront/API Gateway. Managed rules + custom rules
Azure Azure WAF (Front Door / App Gateway) OWASP CRS 3.2 included by default. Policy-based management
Google Cloud Cloud Armor Integrated DDoS + WAF. Preconfigured WAF rules + adaptive protection (ML)
OCI OCI WAF Integrates with Load Balancer/Edge. Provides OWASP rule sets

The OWASP Top 10 is an industry-standard summary of the most common web application security threats. Each vendor’s WAF provides managed rule sets that address these threats.

Below is an excerpt of OWASP Top 10 items that can be mitigated with WAF rules:

OWASP Top 10 threat WAF rule response Vendor managed rules
A01 — Broken Access Control Blocks path traversal, forced browsing AWS Managed Rules (Core), Azure CRS, Cloud Armor preconfigured rules
A03 — Injection (SQL/XSS) Pattern matching for SQL injection, XSS AWS SQLi/XSS Rule Group, Azure CRS, Cloud Armor sqli-v33-stable
A05 — Security Misconfiguration Blocks known vulnerable paths AWS Known Bad Inputs, Azure CRS
A06 — Vulnerable Components Blocks known CVE exploits AWS Managed Rules (CVE), Azure Bot Manager
A07 — Authentication Failures Blocks brute force, credential stuffing AWS Account Takeover Prevention, Azure Rate Limiting
Aspect Managed rules Custom rules
Ownership Updated by the vendor or a security partner Written and maintained directly by the user
Best for Baseline OWASP Top 10 defense, rapid deployment Application-specific logic, business rules
Updates Automatically updated by the vendor when new threats are found Updated directly by the user
Cost Charged per rule group (AWS), included by default (Azure/Google Cloud/OCI) Charged by rule count

Practical recommendations:

  • Step 1 — apply managed OWASP rule sets first (start in Count mode to check for false positives, then switch to Block)
  • Step 2 — add application-specific custom rules (protecting specific API paths, geo-based blocking, etc.)
  • Step 3 — mitigate DDoS/brute force with rate-limiting rules
  • Logging — store WAF logs in S3/Log Analytics/Cloud Logging to analyze attack patterns

Threat detection (GuardDuty, Defender, SCC, Cloud Guard) and container/runtime security (Inspector, Defender for Containers, etc.) are covered in detail in Security Posture Management.

  • Switching WAF straight to Block mode — switching to blocking mode without checking for false positives, causing legitimate traffic to be blocked. Verify in Count mode first
  • Adopting BYOK when vendor-managed keys would suffice — choosing self-managed keys without a regulatory requirement, only increasing operational complexity and failure risk
  • Accessing managed services without a VPC Endpoint — accessing S3, KMS, etc. via a NAT Gateway, resulting in unnecessary internet exposure and cost
  • Deferring PQC transition until “after quantum computers arrive” — already exposed to Harvest Now, Decrypt Later attacks. Long-retention data needs hybrid mode applied starting now
  • Is encryption at rest enabled for all storage/databases (verify default encryption)?
  • Have WAF managed rules been applied in Count mode first, with false positives confirmed before switching to Block?
  • Is access to managed services (S3, KMS, etc.) done via VPC Endpoint / Private Link?
  • Have you inventoried the encryption algorithms in use and established a PQC transition roadmap?
  • Have you evaluated confidential computing for sensitive AI inference workloads?

Post-Quantum Cryptography / Confidential Computing

Section titled “Post-Quantum Cryptography / Confidential Computing”