Take an invented Pune start-up, Tarang Health, that moves its appointment app to a public cloud. The founders sleep better: the provider has guards, biometric doors, redundant power, certifications and engineers who do nothing else. Three months later a journalist emails them a link. It opens a storage bucket full of patient lab reports, no login required. Who failed? The provider, whose data centre was spotless, or Tarang, whose engineer ticked one wrong box? The answer is the whole point of the shared responsibility model.

Where the line sits

Every major provider publishes some version of the same idea. The provider is responsible for the security of the cloud: physical facilities, the hardware, the virtualisation layer, and the core services’ underlying software. The customer is responsible for security in the cloud: what they put there, how they configure it, who they let in.

The line moves with the service model:

  • IaaS (virtual machines, networks, storage): the customer manages the guest operating system, patching, firewall rules, applications, identities and data.
  • PaaS (managed databases, app platforms): the provider takes over the operating system and runtime patching; the customer still owns configuration, access and data.
  • SaaS (hosted business software): the provider runs nearly everything, yet the customer still owns user accounts, permissions, sharing settings and the data they enter.

Whatever the model, identity and data remain with the customer. That sentence is worth underlining, because most cloud breaches live there.

What went wrong at Tarang

A developer needed a partner lab’s software to read reports. Under deadline, they set the bucket’s access policy to allow anyone to read, planning to tighten it later. Later never came. Storage services have long offered controls to block public access at account level; many have turned the safe setting on by default for new buckets in recent years. Tarang’s problem was that an explicit policy overrode caution, and nothing checked it.

Notice the shape: no hacker, no malware, no vulnerability in the provider’s code. A permitted configuration used wrongly. Also notice that the cloud makes this easy: one setting, changed in seconds, effective worldwide. In a private data centre, exposing a server to the internet required a firewall change someone would review.

The other big family: identity and access mistakes

In cloud environments, identity is the new perimeter. Typical problems:

  • Roles with wildcard permissions (“allow all actions on all resources”) because it was quicker than working out the exact set.
  • Long-lived access keys created for a script, copied into a repository or a chat message, and never rotated.
  • The root or owner account used for daily work, without multi-factor authentication.
  • Compute instances given far more permissions than their job needs, so a single application flaw becomes access to the whole account.
  • Forgotten accounts of former staff or contractors.

The guiding principle is least privilege: each person and each workload gets only what it needs, preferably through short-lived credentials issued to a role rather than permanent keys.

How teams catch it: posture management

Cloud security posture management, CSPM, is the discipline and the tooling that continuously compares your actual configuration with a desired baseline. It reads your accounts through the provider’s APIs and flags drift: a public bucket, an unencrypted database, a security group open to the whole internet on an administration port, a user without MFA. Many tools map findings to frameworks such as CIS benchmarks.

Tarang would have benefited from three layers:

  1. Guardrails that prevent. Organisation-level policies that forbid public storage for certain accounts, so the mistake cannot be made.
  2. Infrastructure defined as code. Configurations kept in version control and reviewed like any other change, with automated checks before deployment.
  3. Continuous detection. Posture scanning and alerting, with an owner who answers within hours, not weeks.

Prevention is better than detection, because detection means the data was exposed for some period, and search engines and scanners that index open storage can be quick.

Reading the provider’s documents properly

Provider compliance reports and certifications, such as ISO 27001 or SOC 2 attestations, cover the provider’s part. They do not make your workload compliant. Your auditors will ask for your evidence on access reviews, encryption choices, logging and backups. If you handle personal data, your obligations under Indian law remain yours, wherever the servers sit.

For Tarang, the bucket was locked within minutes of the email. The harder tasks followed: finding out from access logs whether anyone else had read the files (only if logging had been enabled beforehand, which it had not), deciding whom to notify, and rebuilding the habits that allowed a temporary shortcut to become permanent.

Figures and names in the examples are illustrative.

Leave a Reply

Trending