Skip to content
Tech Interview Prep
Technical interview guide

The Shared Responsibility Model

What the cloud provider secures versus what the customer is responsible for, and why most cloud breaches fall on the customer's side.

Read
45 min
Practice MCQs
25
Interview Q&A
25
Edition
v4
Editorial status
Reviewed

Scope: AWS, Microsoft Azure, Google Cloud, and NIST shared-responsibility guidance current 2026-09-01.

Curated: · Written: · Reviewed:

25 questions

MCQ-16A payments team moves its Postgres cluster to a managed database service. The provider now runs the hosts, patches the engine, and takes automated backups. In a design review the team argues that because the provider operates those controls, the team no longer owns any of them. Which responsibility does the team still carry?
MCQ-11An architecture review must draw one responsibility boundary that holds whether the workload runs on IaaS virtual machines, a managed PaaS runtime, or a SaaS CRM. Which layer does the provider secure on its own in all three?
MCQ-6A team is sorting controls for an IaaS workload into security 'of' the cloud and security 'in' the cloud. Its only access is through the provider API and its own guest instances. Which domain belongs on the provider's side of that split?
MCQ-2A team runs its API on a managed Kubernetes service. The provider operates the control plane (API server, etcd); the team runs its own worker node groups. A vulnerability scan reports a remotely exploitable CVE in the container runtime on the worker nodes, whose node OS image is 14 months old. The control plane is on a supported version, and the provider's status page shows no open incidents. Which action must the team take to remediate the finding?
MCQ-3A team runs 30 Ubuntu 22.04 VMs on a standard IaaS provider. The provider patches the hypervisor and firmware. The team has enabled the provider's managed OS patching service for all 30 VMs. A critical OpenSSL CVE is announced. Which action matches the shared responsibility model for these VMs?
MCQ-7A team runs a PostgreSQL database on Amazon RDS. The instance has automated backups, Multi-AZ, and storage autoscaling enabled, and the platform team owns the parameter group and CloudWatch alarms. Which task remains the application team's responsibility?
MCQ-22A team runs a Node.js service on a managed PaaS. The provider patches the runtime and OS, terminates TLS, and manages the network; the team owns the app, its dependencies, and its data handling. A scanner reports that the app's `package-lock.json` pins `jsonwebtoken@8.5.1`, which accepts `alg: none` when the verification key is a string, and the app verifies JWTs with a shared secret. The provider's runtime is fully patched. Which action closes the finding?
MCQ-1A team moves its application off 40 self-managed Ubuntu 22.04 VMs and onto its provider's serverless container service. The provider now patches the host OS and the runtime, while the team keeps building the container image itself in CI from the ubuntu:22.04 base, pinned by digest, with CI rebuilding only when a commit lands on main. A vulnerability is announced in a library bundled in that base image. Which action falls on the team's side of the responsibility split?
MCQ-9A team runs its API as containers on a managed Kubernetes service. The provider manages the control plane and the worker node operating system, including kernel patching, on a 10-day cadence. The team supplies the container images and the pod IAM roles. A critical CVE is published in the OpenSSL library, which is present both in the node OS and in the team's container image. The provider's advisory states that node OS patching will land in its next maintenance window, 10 days out, and that container images are the customer's responsibility. The team's policy requires critical fixes within 4 days. Which action meets the policy?
MCQ-13A team's web app runs on instances behind a provider-managed load balancer in a VPC. The load balancer's frontend listener is HTTPS on 443. A packet capture on an app instance shows the load balancer opening a plaintext HTTP connection to port 8080 on the instance. The load balancer and the instances are in the same VPC, on provider-operated hardware. An auditor asks who is responsible for encrypting the load-balancer-to-instance hop and what change closes the finding.
MCQ-8A payments team keeps 40 million cardholder rows in a managed Postgres service. The provider encrypts every volume at rest with keys it generates and rotates; the team never holds them. A post-incident review finds the internal support tool authenticates with one shared service account and can fetch any row by customer ID; last quarter 12 support staff ran 3,400 such lookups, and the database logs only connection events, not rows read. The compliance lead says the team's only remaining duty is to keep the provider's encryption attestation on file. Which change actually closes the gap the provider's encryption leaves open?
MCQ-4A team migrates a self-managed PostgreSQL 16 cluster on EC2 to Amazon RDS for PostgreSQL. The pre-migration runbook lists five recurring tasks: apply OS security patches monthly, apply PostgreSQL minor upgrades quarterly, take nightly logical backups, rotate database credentials every 90 days, and review PII column access grants for 40 tables quarterly. After cutover, an auditor asks for evidence that OS patches were applied in the last 90 days and that the PII access grants were reviewed this quarter. Which response is correct?
MCQ-5A payments team is assembling evidence for its SOC 2 audit. Its card-tokenization vendor supplies a SOC 2 Type II report covering the vendor's tokenization service for the previous 12 months. What does that report establish for the team?
MCQ-15A bank runs card processing on a hyperscaler and buys a managed log service from a third-party publisher listed in that provider's marketplace. The publisher's agent runs in the bank's account and ships events to the publisher's SaaS control plane. The provider's SOC 2 Type II report is current; its system description covers the provider's own compute, storage and network. Which responsibility map should the bank file for the third-party risk review?
MCQ-25A payments team is completing a PCI DSS 4.0 self-assessment. Their evidence file marks 'disk encryption on the tokenization service' as an inherited control, citing the vendor's public documentation, which states AES-256 encryption at rest. The signed contract's security exhibit says the vendor will 'maintain administrative, physical, and technical safeguards consistent with industry practice' and does not mention encryption at rest. The MSA's change-order clause requires written amendments signed by both parties. The vendor's account team emails that encryption at rest is enabled on the team's tenant. The assessor reviews in two weeks. What should the team do?
MCQ-24A payments team runs its service on a managed Kubernetes platform. The shared-responsibility register lists 140 controls: 56 the provider owns under the contract and 84 the customer owns. An audit of the 84 customer-owned controls finds 27 with no named owner, 35 whose evidence is older than the 90-day attestation window, and 11 that are both unowned and stale. The register's bar is a named owner plus evidence no older than 90 days. How many customer-owned controls currently fail that bar?
MCQ-18A UK bank runs its card-processing service on a public IaaS provider. The provider's evidence pack holds an ISO 27001 certificate whose scope statement names only its data centres and hypervisor, plus a SOC 2 Type II report for 1 March 2024 to 28 February 2025 that lists 11 complementary user entity controls; no earlier attestation report exists. The bank's 40-entry control matrix tags every entry as provider-covered, and the regulator has asked for evidence covering the 2024 calendar year. Which response is defensible?
MCQ-19A team runs 40 production workloads on a managed Kubernetes platform with four node pools: three platform-managed, and one self-managed bare-metal pool the team runs and joins to the cluster itself. The provider's 2024-11 changelog says node OS patching is now automatic for platform-managed pools, control-plane etcd is now encrypted with a provider-managed key (existing customer-managed keys were migrated in that release), and the customer-managed key option is deprecated with removal on 2025-05-31. The team's shared-responsibility matrix has one row per control with a single owner column naming the party that operates it, and two rows still name the customer: 'patch node OS monthly' and 'customer-managed KMS key for control-plane etcd'. Those rows feed the control claims in the team's SOC 2 evidence pack, and the runbook patches all four pools on the first Tuesday of every month. It is 2024-12. Which edit set makes the matrix and the runbook describe what the team actually operates?
MCQ-20A team runs 240 workloads on a cloud IaaS. An audit found 36 storage buckets publicly readable. The provider then enabled an account-level default that blocks public access on every bucket and an org policy that denies creation of any new public bucket. The team's static-site pipeline, which creates two public buckets per release, now fails at bucket creation with an access-denied error, and the security dashboard reports zero public buckets. Which statement correctly describes what changed and what the team must do?
MCQ-12A CI runner in subnet 10.0.4.0/24 assumes three roles in the same account: `deploy`, `read-only`, and `audit`. At 02:14 UTC, `AssumeRole` for `deploy` starts returning `AccessDenied` on every attempt, while `read-only` and `audit` keep assuming successfully from the same runner. Sessions issued before 02:14 continue to work, and the provider's status page reports no IAM or STS degradation. Which explanation fits, and whose configuration must change?
MCQ-10A team runs a three-tier web app in us-east-1: an Application Load Balancer enabled in us-east-1a and us-east-1b, an Auto Scaling group of six m5.large instances, and a Multi-AZ RDS for PostgreSQL instance. The ASG is configured with a single subnet, us-east-1a, and the target group registers those instances. During an AZ-level power event in us-east-1a, RDS failed over to us-east-1b in 70 seconds, but the ALB reported zero healthy targets and the app returned 503s for 19 minutes until the AZ came back. Which change would let the app keep serving through the next single-AZ event?
MCQ-17A team runs its orders database on a managed PostgreSQL service. The provider's SLA guarantees 99.95% instance availability and replicates storage across three AZs. The console shows automated daily snapshots taken at 03:00 UTC with 7-day retention; point-in-time recovery is not enabled. At 14:02 on Tuesday a migration script drops the orders table, and the team's RPO for this database is 5 minutes. Which conclusion about the recovery is correct?
MCQ-23A team is exiting a managed analytics provider. The contract ends on day 0. The provider's termination policy: the export endpoint is available during the contract and for 14 days after contract end, then closes; encrypted backups are purged on day 35; a deletion certificate is issued only if requested before the purge. The team must move 4.2 TB (4200 GB) of event data to its own object store. Measured export throughput is 12 GB/hour and the export runs continuously. The draft runbook says: 'Start the export on day 0 and let the provider's backups cover anything that doesn't finish.' Which revision to the runbook is correct?
MCQ-21A bank's contract with its SaaS vendor requires notification of a confirmed security incident within 24 hours. The vendor's published playbook notifies customers within 72 hours of containment. Over the last year the vendor confirmed six breaches and the bank learned of them 22, 30, 41, 55, 63 and 96 hours after confirmation. The bank's regulator requires it to notify affected customers within 72 hours of becoming aware. Which entry belongs in the bank's risk register?
MCQ-14Your team runs 40 topics on a managed streaming platform. The provider discloses a deserialization flaw: a crafted record published by any tenant can execute code in the broker process. The provider's patched broker image reaches your region in 72 hours and the rollout cannot be accelerated. The advisory's only customer-side mitigation is `record.validation=strict` on a topic, which rejects crafted records but also drops any record whose schema is not registered. Nine of your topics ingest unregistered records from a partner feed; the other 31 do not. The partner feed can replay anything sent while it is paused, but records dropped by validation are unrecoverable. What should your team do?