Why cloud security is different

What "the cloud" really is, the three service models, and where things go wrong

By the end of this lesson you can

  • Define cloud computing using the five essential characteristics from NIST SP 800-145.
  • Tell IaaS, PaaS, and SaaS apart, with one example of each, and name the four deployment models.
  • List the most common causes of cloud security problems, starting with misconfiguration and identity.
  • Explain why an API-driven control plane changes the attack surface compared with a server in a back office.

01Lakeside Bakes moves to the cloud

Meet the example that runs through this whole course. Lakeside Bakes is a small, fictional bakery. Alice owns it. Sam, a part-time developer, built its order portal at orders.example.com, which today runs on an old computer in the back office. The computer is noisy, nobody remembers to back it up, and last winter it stopped for two days when the power failed.

Alice wants to move the portal to a public cloud. Sam is enthusiastic: no hardware to buy, a server ready in minutes, backups as a setting. Both are right. But some things change, and some do not.

What does not change: the order portal still has code with bugs, still stores customer names and addresses, still needs passwords, still needs updates. A SQL injection flaw is just as dangerous on a cloud server as on the back-office computer.

What changes: the "building" is now a website and an API. Anyone with Alice's cloud password, or with an access key Sam leaves in a file, can create servers, copy data, or delete everything, from anywhere in the world, in seconds. A storage setting that used to be impossible to get wrong (the back-office disk was simply not on the internet) can now be switched to public with one click. And every resource costs money while it exists, whether anyone remembers it or not.

This course is about those changes. You will not need to spend money: most labs run on your own computer.

02What "the cloud" is

"The cloud" is a marketing word, so security people use a precise definition. The most cited one comes from the US National Institute of Standards and Technology, in NIST SP 800-145 (final since September 2011 and still current). In plain words, cloud computing is a way of getting on-demand network access to a shared pool of computing resources (servers, storage, networks, applications) that can be set up and released quickly, with very little effort from the provider's staff.

NIST names five essential characteristics. If one is missing, it is not really cloud in the NIST sense:

1

On-demand self-service

You create resources yourself, without calling anyone. Example: Sam starts a server for the portal at 23:00 on a Sunday.

2

Broad network access

Everything is reached over the network with standard tools: browser, command line, apps. Example: Alice checks the account from her phone.

3

Resource pooling

The provider serves many customers from the same hardware, kept apart by software. Example: Lakeside's server shares a physical machine with strangers.

4

Rapid elasticity

Capacity grows and shrinks quickly, sometimes automatically. Example: more capacity for the week before a holiday, less afterwards.

5

Measured service

Use is metered and billed. Example: the bill counts storage, traffic, and server hours, including the test server Sam forgot.

Each characteristic has a security side. Self-service means anyone with the right login can create (and forget) resources. Broad network access means the management interface is on the internet. Pooling means isolation between customers is the provider's job, and your configuration decides what you expose. Elasticity means a stolen key can start a lot of machines very fast. Measured service means mistakes show up on the bill, which is why lesson 2 starts with a budget alarm.

03Service models and deployment models

NIST also defines three service models. They differ in how much of the technology stack the provider runs for you:

  • IaaS (Infrastructure as a Service): the provider runs the buildings, hardware, and virtualization. You get a virtual machine and you manage its operating system, software, network rules, and data. Lakeside example: a virtual server running the order portal.
  • PaaS (Platform as a Service): the provider also runs the operating system and the runtime. You deploy code and configure the service. Lakeside example: a managed database for the orders, where Sam never logs in to an operating system.
  • SaaS (Software as a Service): the provider runs the whole application. You manage users, settings, and your data. Lakeside example: web email, or a hosted accounting tool.
A technology stack of seven layers, from facility at the bottom through hardware, virtualization, operating system, runtime, and application to data at the top. Three columns show the split: In IaaS the customer manages data, application, runtime, and operating system. In PaaS the customer manages data and application. In SaaS the customer manages only data, users, and settings. The provider manages everything below. IaaS PaaS SaaS Data, users, settings Application Runtime Operating system Virtualization Hardware Facility YOU PROVIDER YOU PROVIDER YOU PROVIDER
The higher the service model, the more of the stack the provider runs. The top layer, your data, users, and settings, stays yours in every model.

This picture is simplified on purpose. Lesson 3 shows how the providers draw the line in more detail, including areas that are shared. For now, remember the shape: the provider's part grows from left to right, and yours never shrinks to zero.

Four deployment models

NIST also names four deployment models, which describe who the cloud is for:

  • Private cloud: used by one organization only, whether it runs it itself or someone runs it for it.
  • Community cloud: shared by a group of organizations with common needs, for example the same compliance rules.
  • Public cloud: open to the general public, run by a provider. AWS, Microsoft Azure, and Google Cloud are the three this course names.
  • Hybrid cloud: two or more of the above, connected so that data and applications can move between them.

Lakeside will use a public cloud, plus its own laptops and card terminal on premises. That mix is normal, and it means some security work stays in the bakery.

Myth

"The provider secures everything, that is what we pay for." The provider secures its data centers, hardware, and the services it runs. It does not decide who in your team gets admin rights, whether your storage is public, or whether your server is patched in IaaS. Those are settings you choose. Lesson 3 is entirely about this split.

04What goes wrong

The Cloud Security Alliance (CSA), a nonprofit industry group, publishes a ranked list of the threats its members see most. The 2024 edition, released in August 2024, puts them in this order. The right-hand column shows where this course deals with each one.

RankThreat (CSA Top Threats 2024, paraphrased)Where in this course
1Misconfiguration and inadequate change controlLessons 10, 13, 17 to 19
2Identity and access managementLessons 5 to 9
3Insecure interfaces and APIsLessons 4, 14
4Inadequate selection or implementation of cloud security strategyLessons 3, 24
5Insecure third-party resourcesLesson 20
6Insecure software developmentLessons 19, 21
7Accidental cloud disclosureLessons 10, 11
8System vulnerabilitiesLessons 17, 20
9Limited cloud visibility or observabilityLessons 15, 16
10Unauthenticated resource sharingLesson 10
11Advanced persistent threatsLessons 16, 23

Look at the top two. Neither is a clever hacking technique. A misconfiguration is an insecure setting rather than a bug in code: a storage bucket left public, a firewall rule open to the whole internet, logging switched off. Identity problems are the people side of the same coin: too many admins, passwords without MFA, old access keys nobody deactivated. Both are mistakes that defenders can find and fix with ordinary discipline, which is good news for a small team like Lakeside's.

Notice also what is not at the top: exotic attacks on the provider's own hardware. They exist, but the everyday risk for a small customer sits in its own settings and identities.

05A new attack surface: everything is an API

In the back office, changing the order portal meant sitting at the computer. In the cloud, every action, from creating a server to deleting a backup, is a request to the provider's management API. That management layer is called the control plane. The web console, the command-line tool, and automation scripts are all just different ways of sending it requests. You will see this for yourself in lesson 4.

Three consequences matter for security:

  • Keys travel. Programs talk to the control plane with credentials such as access keys. Keys end up in code, configuration files, laptops, and build systems. Whoever copies a key can use it from anywhere, and automated scanners look for leaked keys continuously.
  • Mistakes are instant and wide. One command can make a bucket public or open a port to the world. There is no delay in which a colleague might notice, unless you build one (reviews, scanners, alerts).
  • Cheap to create also means cheap to forget. A test server, a copy of the database, a temporary key: each takes seconds to create and nobody is forced to clean it up. Forgotten resources keep costing money and keep their weaknesses.

The upside is just as real. Because everything is an API, everything can be recorded (audit logs, lesson 15), checked automatically (scanners, lesson 17), and described in reviewable text files (infrastructure as code, lesson 18). A back-office computer gives you none of that for free.

Law and ethics

Practice only on accounts and resources you own, or that you have been explicitly invited to test in writing. Looking for "open" storage or trying leaked keys that belong to someone else is unauthorized access in most countries, even if the resource was misconfigured. Every lab in this course uses your own emulator, your own files or, optionally, your own account.

06How this course works

The order of this course loosely follows the CSA Security Guidance version 5 (July 2024), which organizes cloud security into twelve domains, from cloud concepts and governance, through identity and access management, monitoring, infrastructure and networking, workloads, data and applications, to incident response. You will meet most of those areas in that spirit, at a practical, small-organization scale.

Every lesson follows the same rhythm:

  1. Concept first, explained without any provider's product names.
  2. How the providers name it: a small table for AWS, Azure, and Google Cloud. Names differ. Ideas mostly do not.
  3. What goes wrong: the usual misconfiguration, and how to prevent and detect it.
  4. Hands-on: local tools first (a free AWS emulator called moto, OpenTofu in plan-only mode, the scanners Checkov and Trivy). Optional steps in a real account come second, and only after the budget alarm of lesson 2.
  5. Cleanup, always: every lab ends by removing what it created.

You do not need a cloud account or a credit card for any required step. If you do have one, treat it with respect from the first minute.

Defender's view

Three habits carry through every lesson. Least privilege: every person and program gets only the access it needs. Logs on: keep a record of who did what, and look at it. Budget alarm first: before you create anything real, set up a spending alert, so a mistake or a stolen key shows up as an email rather than as a surprise at the end of the month.

Practical exercise: map Lakeside's services

Paper and pen, or a text file. No account needed. Check off each step as you finish it. Your checkmarks are saved in this browser.

Key takeaways

  • NIST defines cloud computing by five characteristics: on-demand self-service, broad network access, resource pooling, rapid elasticity, and measured service.
  • IaaS, PaaS, and SaaS differ in how much of the stack the provider runs. Your data, users, and settings stay yours in every model.
  • The four deployment models are private, community, public, and hybrid.
  • CSA's Top Threats 2024 puts misconfiguration first and identity and access management second: everyday mistakes, not exotic attacks.
  • The control plane is an internet-facing API, so credentials, quick mistakes, and forgotten resources are the new attack surface.