Cloud Security Architecture

In 2019, Capital One experienced one of the most significant cloud security breaches in history. Over 100 million customer records were exposed—not because AWS's infrastructure was compromised, but because the bank misconfigured a Web Application Firewall. The attacker exploited Capital One's configuration, not Amazon's cloud platform. This distinction is crucial: the cloud provider's security and the customer's security are not the same thing.

This fundamental reality is formalized in what cloud providers call the Shared Responsibility Model—the cornerstone of all cloud security architecture. Without a deep understanding of this model, organizations invariably make one of two costly mistakes: they either assume the cloud provider handles all security (leaving critical gaps), or they attempt to secure everything themselves (wasting resources on areas the provider already secures).

Mastering this model is the first step toward building genuinely secure cloud architectures.

The Shared Responsibility Model is a security and compliance framework that explicitly delineates which security obligations belong to the cloud service provider (CSP) and which belong to the customer. It exists because cloud computing fundamentally changes the security paradigm—you're no longer in full control of the entire stack, but you're still fully accountable for your data and applications.

The fundamental principle is simple:

• Cloud Provider: Security OF the cloud (the infrastructure itself)
• Customer: Security IN the cloud (what you build and deploy)

This seemingly simple division has profound implications. The provider guarantees that their data centers are physically secure, that hypervisors isolate workloads properly, and that the underlying network infrastructure is protected. But they have no control over—and therefore no responsibility for—how you configure your S3 bucket permissions, whether your application validates input properly, or if your employees use weak passwords.

The philosophical foundation:

The model is rooted in a fundamental truth about security: you cannot secure what you cannot control. The cloud provider controls the hardware, the hypervisors, and the physical facilities—so they secure those. You control your data, your applications, and your configurations—so you must secure those.

This isn't a limitation of cloud computing; it's a feature. By clearly separating responsibilities, both parties can focus their security efforts on their areas of expertise and control. AWS knows more about securing hypervisors than you do; you know more about your application's security requirements than AWS does.

Cloud providers invest billions of dollars in securing their infrastructure. Understanding exactly what they protect helps you avoid both gaps (assuming they secure something they don't) and redundancy (trying to secure something they already handle).

Physical Security:

Major cloud providers maintain data centers with security measures that would be economically impossible for most organizations to replicate:

• Biometric access controls — Multi-factor authentication including retinal scans, fingerprints, and security badges
• 24/7 video surveillance — Every angle of every facility continuously monitored and recorded
• Environmental controls — Fire suppression, flood detection, and climate control systems
• Secure hardware disposal — Drives are degaussed, shredded, and destruction is certified
• Undisclosed locations — Data center locations are not publicly advertised to prevent targeted attacks
• Physical intrusion detection — Motion sensors, security patrols, and perimeter defenses

Compliance and Certifications:

Cloud providers undergo rigorous third-party audits and maintain certifications that would be prohibitively expensive for individual organizations to obtain:

These certifications cover the provider's infrastructure. They do not automatically extend to what you build on that infrastructure—a crucial distinction we'll explore in the customer responsibility section.

While cloud providers secure the foundation, customers are responsible for everything they build on that foundation. This is where the majority of cloud security breaches occur—not because the provider failed, but because the customer misconfigured, neglected, or misunderstood their responsibilities.

Customer responsibility spans several critical domains:

The "Shared" in Shared Responsibility:

Some areas involve overlapping responsibilities where both provider and customer must act:

• Encryption — The provider offers encryption capabilities and may manage key infrastructure (like AWS KMS), but you must enable encryption and may need to manage keys depending on your requirements.

• Patch Management — For managed services, the provider patches the underlying infrastructure. For IaaS, you patch the OS. For PaaS, the division depends on the service.

• Network Security — The provider secures the physical network and offers tools like VPCs and security groups. You must configure those tools to implement your security requirements.

• Logging and Monitoring — The provider offers logging services (CloudTrail, CloudWatch, etc.). You must enable them, configure appropriate alerts, and actually review the logs.

These shared areas are often where confusion arises and breaches occur—organizations assume the provider is handling something that actually requires customer action.

The shared responsibility model isn't static—it shifts dramatically depending on the cloud service model you're using. As you move from IaaS to PaaS to SaaS, more responsibility transfers to the provider, but you never escape all customer responsibilities.

The fundamental principle holds: the more the provider manages, the less you manage, but you always remain responsible for your data and how you configure access to it.

IaaS: Maximum Control, Maximum Responsibility

With Infrastructure as a Service (EC2, Azure VMs, GCE), you get virtual machines and networking. Everything above the hypervisor is your responsibility:

• You choose, install, and patch the operating system
• You configure firewalls and security groups
• You install and secure all middleware and runtimes
• You deploy and secure your applications
• You manage all data encryption

This model offers maximum flexibility but requires the most security expertise. You're essentially running a traditional data center, just on rented hardware.

PaaS: Reduced Scope, Shifted Responsibility

With Platform as a Service (Lambda, App Engine, Azure Functions), the provider manages the runtime:

• The provider patches and secures the operating system
• The provider manages the execution environment
• You write and deploy code, but in a constrained environment
• You still manage application logic and security
• You control access and data handling

PaaS reduces your attack surface but doesn't eliminate security concerns. Your code can still have vulnerabilities, your IAM can still be misconfigured, and your data can still be exposed through application logic errors.

SaaS: Minimal Infrastructure, Persistent Data Responsibility

With Software as a Service (Salesforce, Workday, Slack), the provider manages nearly everything:

• The entire application stack is provider-managed
• You don't deploy code or manage infrastructure
• You still configure access controls and user permissions
• You still classify data and decide what to store
• You still ensure your users follow security practices

Even with SaaS, you can't outsource accountability. If you give an intern admin access, or export sensitive data to insecure locations, or fail to offboard terminated employees, those are your failures.

While the fundamental concept is universal, each major cloud provider articulates the shared responsibility model with slightly different emphasis and terminology. Understanding these nuances is essential when working in multi-cloud or hybrid environments.

Google Cloud Platform (GCP):

GCP frames shared responsibility with additional emphasis on Google's infrastructure investments and Fate Sharing—the concept that Google's business depends on customer security, so they actively invest in tools that help customers secure their workloads:

• Strong defaults and guardrails built into services
• Security Command Center for unified visibility
• Emphasis on secure-by-default configurations
• Confidential Computing initiatives for enhanced workload protection

Key Differences to Note:

While these differences exist, the core model remains consistent: providers secure the infrastructure, customers secure their workloads. Any differences are in implementation details, not the fundamental division of responsibility.

Understanding the model conceptually isn't enough—you must operationalize it within your organization. This requires systematic processes, clear ownership, and continuous validation.

Step 1: Create a Responsibility Matrix

For every cloud service you use, document exactly which security controls are provider responsibilities and which are yours. This matrix becomes your source of truth:

Service: Amazon S3

| Control              | Provider | Customer | Notes                              |
|---------------------|----------|----------|------------------------------------|
| Physical disk security | ✓      |          | Provider manages hardware          |
| Encryption at rest  |          | ✓        | Customer must enable SSE           |
| Bucket policies     |          | ✓        | Customer defines access rules      |
| Access logging      |          | ✓        | Customer enables CloudTrail        |
| Object tagging      |          | ✓        | Customer classifies data           |
| SSL/TLS in transit  | ✓        | ✓        | Provider offers, customer uses     |
| Versioning          |          | ✓        | Customer enables for recovery      |

Create this matrix for every service in your architecture. It forces you to explicitly address each security dimension.

Even organizations that understand the shared responsibility model often fall into predictable traps. Awareness of these pitfalls can prevent your organization from becoming another breach statistic.

The Abstraction Trap:

Higher abstraction levels (PaaS, serverless) can create false confidence. When you can't see the operating system, it's easy to forget that security still requires attention. Consider a Lambda function:

• You don't manage the OS (provider handles)
• You don't patch runtime (provider handles)
• But your code can still:
  • Accept unsanitized input leading to injection attacks
  • Log sensitive data to CloudWatch
  • Have overly permissive IAM roles
  • Connect to resources without encryption
  • Expose secrets through environment variables incorrectly

The attack surface shrinks with abstraction, but it never disappears entirely. Your responsibilities change form but don't vanish.

The Shared Responsibility Model is the conceptual foundation upon which all cloud security architecture is built. Without internalizing this model, organizations cannot effectively secure their cloud deployments.

What's Next:

With the shared responsibility model as our foundation, we'll explore the first major customer responsibility in detail: Identity and Access Management (IAM). IAM is where most cloud security implementations begin, and where many security failures originate. Understanding IAM deeply is essential for fulfilling your side of the shared responsibility bargain.