0%
Security Engineering for Software Teams
Thinking About Security
Cryptography You Will Actually Use
Identity
Vulnerability Classes
Detection and Response
Cloud Misconfiguration and Least Privilege
Most cloud security incidents do not involve a clever exploit of the cloud provider. They involve customers configuring their own resources insecurely: a storage bucket left public, a database snapshot shared with everyone, an access key with administrator rights committed to a repository or a security group open to the whole internet. This section explains why misconfiguration dominates cloud incidents and what makes cloud environments prone to it.
The shared responsibility model
Cloud providers secure the underlying infrastructure: data centres, hardware, hypervisors and the managed service software. Customers secure what they build on top: identity and access configuration, network settings, data protection, operating systems on virtual machines and application code. The split varies by service, with managed services taking on more of the work, but configuration of access and exposure always stays with the customer. Most incidents fall on the customer side of the line.
Why misconfiguration is so common
Cloud platforms expose hundreds of services, each with dozens of settings that affect security. A single storage service may have access policies, access control lists, public access blocks, encryption settings, logging and replication rules, any of which can expose data. Teams provision resources quickly, often through consoles or copied templates, and security relevant settings are easy to overlook. Defaults have improved, but older resources and permissive templates remain.
Speed and self service
Cloud allows any engineer with the right permissions to create infrastructure in minutes. This speed is a major benefit, but it means security reviews that once happened before servers were bought no longer happen by default. Without guardrails, one engineer can create a public database endpoint or grant a role administrator rights in seconds, and nobody notices until something goes wrong.
The control plane is the new perimeter
In the cloud, the control plane API is how everything is created, changed and destroyed. Anyone with suitable credentials can read data, change network rules or delete backups through API calls from anywhere on the internet. This makes cloud identity and access management the most important security control. A leaked access key with broad permissions is equivalent to physical access to the entire data centre.
Public storage exposure
Publicly readable storage buckets have caused many data leaks, exposing backups, logs, customer exports and internal documents. They usually result from a policy granting read access to everyone, often set to share files with a partner or to host static content, then later used for sensitive data. Automated scanners on the internet find such buckets quickly, so exposure is rarely unnoticed by attackers for long.
Snapshots, images and shared resources
Beyond buckets, many resources can be shared publicly or with other accounts: database snapshots, disk snapshots, machine images, container registries and message queues. A snapshot of a production database shared publicly exposes the whole database. Teams sometimes share resources across accounts for migration or testing and forget to remove the share afterwards.
Network exposure
Security groups and firewall rules open to the whole internet on SSH, remote desktop or database ports invite brute force attempts and exploitation of unpatched services. These rules often start as temporary troubleshooting changes that are never removed. Management interfaces of managed services, such as search clusters or caches, have also been exposed publicly with no authentication.
Credential leaks
Long lived access keys are frequently leaked through code repositories, logs, container images and laptops. Attackers scan public repositories for key patterns and use them within minutes. The impact depends on the permissions attached to the key: a key scoped to one bucket leaks one bucket, while a key with administrator rights leaks the whole account.
Overly broad permissions
Permissions that grant far more than needed turn small mistakes into large incidents. A web application role allowed to read all buckets, a CI role able to change identity policies or a developer group with administrator rights in production all amplify the damage of any compromise. Broad permissions are common because they are easy to write and make things work quickly.
Compromise chains
Real cloud incidents chain several weaknesses. An SSRF vulnerability lets an attacker query the metadata service, obtain the instance role credentials, use those credentials to list and read buckets because the role is broad, and copy data out because no egress or data access monitoring is in place. Fixing any one link breaks the chain, which is why defence in depth matters in the cloud as elsewhere.
Multi account sprawl
Large organisations run hundreds or thousands of cloud accounts, subscriptions or projects. Each may have its own identities, networks and data. Without central visibility, teams lose track of what exists, who owns it and how it is configured. Forgotten accounts with old resources and broad access are common sources of incidents.
Infrastructure as code helps and hurts
Infrastructure as code makes configuration reviewable, repeatable and testable, which helps security. But it also replicates mistakes: a permissive module used by fifty teams creates fifty exposures. Treating shared modules as security sensitive code, with careful review and secure defaults, multiplies the benefit instead of the risk.
Attackers automate discovery
Attackers continuously scan the internet for open ports, exposed management interfaces and public buckets, and they scan code hosting sites for keys. The time between an exposure appearing and it being found by automated tools is often minutes to hours. Cloud security therefore cannot rely on periodic manual audits; it needs prevention and continuous detection.
Defaults that changed over time
Cloud providers have tightened many defaults, such as blocking public access on new buckets and requiring session tokens for instance metadata. These changes protect new resources but rarely change existing ones. Accounts created years ago may still contain resources configured under old defaults, so teams cannot assume that today's safe defaults describe their whole estate.
Visibility gaps
Many organisations cannot answer simple questions quickly: which buckets are public, which roles have administrator access, which keys are older than a year. Without central visibility, risky resources stay hidden until an attacker or researcher finds them. Building that visibility is often the first and most valuable step in a cloud security programme.
Logging that was never turned on
Many incidents are made worse because the logs needed to investigate them were never enabled. Data access logging on storage, flow logs on networks and audit trails in some regions are often off by default or disabled to save cost. When an exposure is discovered, the team cannot tell whether anyone actually read the data, which can turn a near miss into a reportable breach.
Human error under pressure
Many misconfigurations happen during incidents, migrations or deadlines, when engineers make quick changes to restore service or unblock a launch. A rule opened to the internet to debug a connection, or a role given administrator rights to finish a migration, is meant to be temporary. Without expiry and follow up, temporary changes become permanent exposures.
Most cloud incidents come from customer misconfiguration: broad permissions, public storage, leaked keys and open network rules. The control plane API makes identity the main perimeter. Speed, sprawl and copied templates make mistakes easy, and attackers find them within minutes, so prevention and continuous detection are both required.