Author Archives: Piotr Mech

Beyond “Lift and Shift”: Untangling Cloud Accountability Across IaaS, PaaS, and SaaS

Reading Time: 3 minutes

For decades, traditional IT security had clear perimeters: corporate data lived behind a physical firewall, on hardware owned and managed by the organization. Transitioning to the cloud has completely upended this model.

Yet, a dangerous misconception persists in many boardrooms and management meetings: the belief that moving to the cloud means outsourcing security.

When an organization migrates workloads to a cloud service provider (CSP), executives often assume that the vendor takes care of cyber defense end-to-end. In reality, cloud security is not outsourced; it is shared. Understanding where the boundary lies—and how it shifts across Infrastructure as a Service (IaaS), Platform as a Service (PaaS), and Software as a Service (SaaS)—is one of the most critical governance challenges facing modern managers.

To map these boundaries effectively, organizations frequently reference frameworks like the UK National Cyber Security Centre’s (NCSC) guidance on the Cloud security shared responsibility model.

The Shared Responsibility Spectrum

The core principle of the shared responsibility model is straightforward: the cloud provider is responsible for security of the cloud (the physical datacenters, hardware, hosts, and underlying virtualization layers), while the customer is always responsible for security in the cloud (their data, identities, configurations, and workloads).

However, the exact dividing line moves dynamically depending on the service model you choose:

  • Infrastructure as a Service (IaaS): In an IaaS model (such as renting raw virtual machines or cloud storage), the customer retains the heaviest operational burden. While the vendor secures the physical host and hypervisor, you are entirely responsible for the operating system (OS) installation, patching, network firewall configurations, middleware, and application security. Treating IaaS like traditional on-premises hardware without updating OS patches or hardening endpoints invites immediate compromise.
  • Platform as a Service (PaaS): PaaS abstracts away much of the underlying infrastructure. The vendor manages the OS, runtime environments, and hardware patching, allowing development teams to focus purely on code. But this reduction in infrastructure burden shifts accountability squarely onto configuration and identity. If a manager allows developers to deploy code with overly permissive API keys or misconfigured database access rules, the platform itself remains secure while the data leaks directly through customer error.
  • Software as a Service (SaaS): In SaaS (such as enterprise email suites, CRMs, or collaboration tools), the vendor handles the vast majority of the stack, including the application code, network controls, and infrastructure. It is tempting for managers to assume they have zero security duties here. Yet, SaaS customers remain entirely accountable for user access governance, data classification, permissions, and tenant configurations. Most modern SaaS breaches do not stem from provider code vulnerabilities; they happen because a user account lacked multi-factor authentication (MFA) or shared a sensitive file publicly.

Reframing Accountability for Managers

For management students and future business leaders, the takeaway is clear: you can delegate operational tasks, but you can never delegate ultimate accountability.

When selecting and deploying cloud services, managers must avoid passive adoption and instead enforce a structured governance approach:

  1. Map Control Ownership Early: Never assume a security control (such as logging, key management, or identity verification) is automatically handled by the vendor. For every cloud service adopted, require engineering and security teams to explicitly document a responsibility matrix.
  1. Treat Configuration as a Risk Vector: As you move from IaaS toward SaaS, infrastructure vulnerabilities decrease, but human configuration errors become the primary threat vector. Accountability must shift toward enforcing strict Identity and Access Management (IAM), Least Privilege access principles, and automated posture management tools.
  1. Align Procurement with Security Clarity: During vendor selection, treat vague service descriptions regarding data protection, logging visibility, and administrative boundaries as unresolved business risks. If a contract does not clearly delineate who monitors logs or manages encryption keys, your organization is walking blind into a compliance gap.

Cloud computing offers unprecedented agility, scalability, and resilience—but only if management treats security boundaries as an active operational partnership rather than an excuse to hand over the keys.

AI Engine – Google Gemini 3

  • Prompt used:

Write a short blog post for management students explaining how security responsibilities differ across IaaS, PaaS, and SaaS. Use and link to the UK National Cyber Security Centre’s “Cloud security shared responsibility model” (https://www.ncsc.gov.uk/collection/cloud/understanding-cloud-services/cloud-security-shared-responsibility-model). Challenge the idea that cloud security is simply outsourced. Focus on how managers should assign accountability when selecting services and configuring them. Don’t invent facts. After the post, include a section for the AI engine and the exact prompt used.

Who Can Touch Your Data in the Cloud? Reading Provider Security Pages Critically

Reading Time: 4 minutes

Moving customer data to a cloud provider doesn’t remove the question of who can see it. It moves the question to someone else’s staff, processes and paperwork. Both Microsoft and Google publish material on how their people are kept away from customer data. This post compares what two of those pages say about employee access, approvals and data separation, and what a manager should still ask.

Two documents, two different jobs

Microsoft Learn’s Azure customer data protection is a short page about how Azure protects customer data. When retrieved, it showed a last-updated date of 24 July 2026.

Google Cloud’s Google infrastructure security design overview is a long design document aimed at security executives, architects and auditors. It covers Google’s infrastructure across consumer services, Workspace and Google Cloud. It states that it was last updated in June 2024 and reflects the position at the time of writing.

So this is not a like-for-like comparison. A topic missing from one page doesn’t prove a control is missing, and a longer page doesn’t prove stronger controls.

Employee access

Microsoft says its operations and support staff can’t access customer data by default. When a support case needs access, it describes a just-in-time model under policies that Microsoft says it audits and vets. The stated policy is no default access, no accounts on customer virtual machines, and least privilege with audited and logged requests. Support staff get unique corporate accounts, multifactor authentication and access only from secure consoles.

Google describes insider-risk measures: limited and monitored administrative access, automation to reduce the need for privileged access, and debugging interfaces designed to avoid exposing sensitive data. Engineers get individual identities, employee access to end-user information can be logged, and a security team says it monitors access patterns. Employee devices are managed, and security keys replace one-time passcodes.

A critical reading: “no access by default” is a default, not a guarantee. Both providers describe legitimate exceptions, such as support cases and outage repair, so the real question is who decides when an exception applies.

Approvals

Microsoft’s page mentions just-in-time access but doesn’t say who approves a request, how long access lasts, or whether the customer is asked or told. It may document that elsewhere. That gap is a question to put to Microsoft, not evidence of a missing control.

Google describes two layers. Internally, it describes approval chains, logging and notification, including a two-person rule so that one engineer can’t perform certain sensitive operations alone. For customers, it describes Access Approval, which it says requires your explicit, cryptographically verified approval before Customer Care or engineering access your data, and Access Transparency logs, which let you check why access occurred. The page doesn’t say whether these are on by default, which services or regions they cover, or what they cost. Note also that the two-person rule applies to certain sensitive actions, not all access.

Data separation

Both providers describe multi-tenant infrastructure, where different customers’ data sits on shared systems.

Microsoft says it uses logical isolation to keep each customer’s data apart and describes the prevention of cross-customer access as rigorous. The page gives no detail on how. It also stresses that you are responsible for ensuring your data is encrypted to your own standards, with tools such as Azure Key Vault to help.

Google is more explicit and more candid. It says its infrastructure generally doesn’t put a customer’s data on dedicated machines, except in specific cases such as sole-tenant nodes in Compute Engine. Instead it describes zero-trust between services, service identities, and layered sandboxing, with more layers for riskier workloads. It also describes Confidential Computing options, default encryption at rest with Google-managed keys, and customer-managed keys through Cloud KMS.

A critical reading: Google gives more mechanism and Microsoft more outcome-language. But describing a mechanism is not evidence that it works.

What customers should verify

  1. Independent evidence. Ask for assurance reports and certifications that cover the specific services, regions and time period you use, and read the scope and exclusions.
  2. Contractual force. Check whether the described controls are contractual commitments or just documentation. Google’s own page says its policies and systems may change.
  3. Who approves provider access. Ask in what circumstances provider staff can reach your data, who authorises it, for how long, whether you must approve or are only notified, and whether any emergency override exists.
  4. Visibility. Ask whether access logs are on by default or opt-in, which services they cover, how quickly they arrive, how long they are kept, and whether you can feed them into your own monitoring.
  5. Keys. Ask who holds your encryption keys, which services support customer-managed keys, and what provider staff can still do if you revoke them.
  6. Isolation. Ask which services share infrastructure with other tenants, what dedicated options exist and at what trade-off, and how isolation failures are tested and disclosed.
  7. What the pages don’t discuss. Neither page addresses legal demands for data from authorities. Google’s page mentions some servers hosted in third-party data centres, so ask about third parties too.
  8. Your own side. Microsoft’s page assigns encryption, records retention and e-discovery to the customer. Your own administrators are also part of the access picture.

Bottom line

Treat these pages as the start of due diligence, not the end. The useful question is not “can the provider access my data?” but “under what conditions, who decides, and how would I know?”

AI engine and prompt used

AI engine: Claude Sonnet 5.5

Exact prompt:

Write a short, critical blog post for management students about who can access customer data inside a cloud provider. Use and link to Microsoft Learn, “Azure customer data protection” (https://learn.microsoft.com/en-us/azure/security/fundamentals/protection-customer-data) and Google Cloud, “Google infrastructure security design overview” (https://docs.cloud.google.com/docs/security/infrastructure/design?hl=en). Compare the safeguards described for employee access, approvals, and data separation. Explain what customers should verify with providers, and don’t present provider claims as independently verified. Don’t invent facts. After the post, include a section for the AI engine and the exact prompt used.

Cloud Security Starts with a Building, but Does Not End There

Reading Time: 2 minutes

“The cloud” sounds intangible, yet it depends on physical buildings, servers and electricity. For managers, protecting that infrastructure matters because theft, damage or disruption can affect business operations.

In “AWS Data Centers,” AWS describes four layers of protection. The perimeter uses measures such as guards, fencing and intrusion detection, depending on the location. The infrastructure layer includes backup power, cooling and fire suppression. The data layer restricts access and separates privileges, while the environmental layer includes choosing locations to reduce exposure to flooding, extreme weather and earthquakes. Together, these measures address both unauthorized entry and threats to availability.

Google’s “Google infrastructure security design overview” describes comparable physical layers: vehicle barriers, cameras, biometric identification, metal detectors and laser-based intrusion detection, followed by additional protection and monitoring of server access. Its wider security design extends into hardware verification, service access controls and encryption. Google also says retired storage devices undergo a cleaning process with two independent verifications; devices that fail are destroyed on-site.

The management lesson is that these layers address different risks. Barriers and access checks can reduce opportunities for theft or tampering; backup systems can reduce disruption. However, a secure building cannot, by itself, stop someone using stolen account credentials or correct permissions that expose confidential information. Physical protection is necessary, but it cannot substitute for sound digital access controls and employee practices.

We should also distinguish descriptions from evidence of effectiveness. These pages explain the providers’ own controls. AWS mentions third-party audits, but that statement is not an independently reviewed audit report. Google explicitly dates its overview to June 2024. Neither page alone establishes how consistently every control operates or proves that one provider is safer than the other.

For managers selecting a cloud provider, the practical implication is to request relevant independent assurance reports, examine their scope and exceptions, and assign clear internal responsibility for access permissions and recovery planning. Buying cloud services should change how security responsibilities are managed – not make them disappear.

AI engine used: GPT-6 Astra
Prompt: Write a short blog post for management students about physical security in cloud data centers. Use and link to both sources: AWS, “AWS Data Centers” (https://aws.amazon.com/trust-center/data-center/) and Google Cloud, “Google infrastructure security design overview” (https://docs.cloud.google.com/docs/security/infrastructure/design?hl=en). Explain the layered controls they describe, then critically assess what physical security can and cannot protect against. Distinguish providers’ descriptions of their own controls from independent evidence. End with a practical management implication. Don’t invent facts.

Tagged ,