The half of cloud security your provider will never do for you

Shared responsibility is a clean diagram covering a messy division of labor. The gap is almost always configuration, identity, and data.

Every major cloud provider publishes a shared responsibility model, and every one of them is accurate. They are also, in practice, one of the most consistently misread documents in enterprise technology. The diagram is not wrong. The trouble is what a reasonable person infers from the phrase the provider handles security.

The provider does handle security. Specifically: physical datacenters, the hypervisor, the host operating system, the network fabric, and the availability of the control plane. That is an enormous amount of security work, performed better than nearly any individual organization could perform it.

It is also not the part where breaches happen.

Where incidents actually originate

Cloud incidents concentrate in a narrow band of causes, and they sit squarely on the customer side of the line:

  • Misconfiguration. A storage bucket set to public. A security group opened to 0.0.0.0/0 during troubleshooting and never closed. A database exposed to the internet because the default in a template was permissive.
  • Identity and access. Over-permissioned roles, long-lived access keys committed to a repository, service principals with subscription-wide rights because scoping them properly was harder.
  • Unprotected data. Sensitive information stored without classification, encryption, or retention rules, often in a service that was stood up for a proof of concept and quietly became production.

None of these are provider failures. Every one of them is fully within the customer's control, and every one of them is invisible without deliberate instrumentation.

The provider guarantees the building. The doors on your floor, the filing cabinets, and who holds the keys: that is the half people underestimate.

The regulatory dimension

GDPR, HIPAA, and GLBA do not recognize the shared responsibility model as a defense. If regulated data is exposed, the accountable party is the organization that held it, not the platform it ran on. A provider's compliance certifications establish that the underlying infrastructure can be used in a compliant way. They say nothing about whether you did.

This distinction becomes concrete during an audit. The auditor is not going to examine the hypervisor. They are going to ask which identities can reach the regulated data set, how access is reviewed, how the data is classified and encrypted, how long it is retained, and how you would know if it were exfiltrated. All customer-side questions.

Closing the gap

The instrumentation that closes this gap is well understood and, increasingly, native to the platforms themselves.

Posture management continuously evaluates configuration against a baseline and flags drift. This is the control that catches the security group somebody opened at 2am during an incident and forgot to close. Microsoft Defender for Cloud and CrowdStrike Falcon Cloud Security both do this well; the important part is that someone owns the findings queue and works it down.

Least-privilege identity in the cloud means scoping service principals and managed identities to specific resources, eliminating long-lived keys in favor of federated or managed credentials, and applying the same time-bound elevation model to cloud administration that you apply on-premises.

Data classification tells you which storage actually matters. Without it, every bucket is equally urgent, which means none of them are. Purview labeling applied at the container level gets you most of the way with a fraction of the effort of per-file classification.

A practical first move

Run a posture assessment against your existing cloud environment and sort the findings by exploitability rather than severity score. Publicly reachable resources with sensitive data and permissive identity attached are the short list. In most environments that list is under twenty items, and closing it takes a couple of sprints.

That is a materially better use of a quarter than a broad hardening program. And unlike a hardening program, it produces a specific, defensible answer the next time somebody asks whether the cloud environment is secure.

Related practice

Cloud & Modern Work

Defender, Purview, and Sentinel deployed to hold up under GDPR, HIPAA, and GLBA, plus Copilot readiness and the Power Platform apps we build and host.