Join our Newsletter — 33% off our NHI Course
Home› Guides› Amazon Inspector, Defender Vulnerability Management and GCP Scanning…
Guide Cloud Security

Amazon Inspector, Defender Vulnerability Management and GCP Scanning Compared: How each cloud scans workloads and where coverage ends

PDF for newsletter subscribers · free to read online

← All guides
By NHI Mgmt Group, edited by Lalit Choda Updated 14 November 2026 16 min read
On this page

Every major cloud now ships its own workload vulnerability scanner, and most teams switch them on without knowing which workloads each one actually reaches. This guide explains how the provider-native scanners on AWS, Google Cloud and Microsoft Azure discover and assess virtual machines, container images, running containers and serverless functions, with agent-based and agentless options side by side. It covers eligibility rules, scan frequency, the identities and permissions each scanner uses, how to find the workloads that fall outside coverage, and how to feed results into one remediation process. It is written for cloud security engineers, vulnerability management leads and platform teams running one or more clouds. Platforms are listed alphabetically; order is never a ranking. For scanning fundamentals (methods, credentialed scans, tuning) start with the Vulnerability Scanning Guide; this page goes deeper on the native cloud services only.

Key takeaways

  • Native scanners are per-workload-type services. Each provider scans VMs, registry images, running containers and serverless code through separate features with separate switches, so "scanning is on" needs checking per workload type.
  • Agentless means snapshot analysis. All three providers read a copy of the disk out of band, on a schedule of roughly 12 to 24 hours, while agent-based scanning reacts to change events; pick per workload, not per estate.
  • Eligibility rules decide coverage. Filesystem types, disk counts and sizes, encryption key placement, image age and pull windows all determine whether a given workload is assessed at all.
  • Scanners are privileged non-human identities. Snapshot and key permissions granted to scanner roles and service agents need the same ownership, scoping and review as any other machine identity.
  • Coverage reports are the control. Measure the gap between the asset inventory and the scanner's coverage list every week, and route findings from all clouds into one prioritised queue.

What the native scanners cover

Provider-native scanning sits inside each cloud's security service family: Amazon Inspector on AWS; Security Command Center (SCC) and Artifact Analysis on Google Cloud; Microsoft Defender for Cloud with Microsoft Defender Vulnerability Management (MDVM) on Azure. The table lists each provider's own feature names for each workload type, from its documentation as of November 2026.

Workload typeAWSGoogle CloudMicrosoft Azure
Virtual machines, agent-basedAmazon Inspector agent-based scanning of SSM managed EC2 instances (Enhanced EC2 Scanning with the Amazon Inspector VM Scanner)VM Manager OS vulnerability reports from the OS Config agent, surfaced in SCC as OS_VULNERABILITYDefender for Servers vulnerability assessment with Defender Vulnerability Management (agent-based)
Virtual machines, agentlessAmazon Inspector agentless method in hybrid scan mode (EBS snapshots)Vulnerability Assessment for Google Cloud (disk clones scanned with SCALIBR)Agentless scanning for machines (disk snapshots analysed by MDVM)
Registry imagesAmazon ECR enhanced scanning by Amazon Inspector (continuous or on push)Artifact Analysis automatic scanning for Artifact Registry Docker repositoriesDefender for Containers registry vulnerability assessment (ACR and connected registries)
Running containersInspector container mapping of ECR images to ECS (including Fargate) and EKS workloadsVulnerability Assessment for GKE Standard and Autopilot containers; Cloud Run (Preview)Defender for Containers runtime vulnerability assessment for AKS and supported Kubernetes
Serverless and codeLambda standard scanning (dependencies) and Lambda code scanning (custom code)Container image findings for Cloud Run services and jobs and App EngineRegistry and runtime image assessment for containerised workloads
Finding categoriesPackage vulnerability, code vulnerability, network reachabilityOS_VULNERABILITY, SOFTWARE_VULNERABILITY, CONTAINER_IMAGE_VULNERABILITYSecurity recommendations per machine and per image

The scanners are posture and workload features, not full replacements for an enterprise scanning programme: they cover the provider's own compute and registries. Anything else (on-premises hosts, SaaS, network devices, unmanaged registries) still needs the tooling described in the Vulnerability Management Guide.

AWS: Amazon Inspector

EC2 instances

  • Scan modes: agent-based scanning uses Systems Manager (SSM) on managed instances and runs continuously, triggered by instance launch, new software installs on Linux and macOS, and new relevant CVEs; SSM inventory is evaluated every 30 minutes by default. Hybrid mode adds the agentless method for EBS-backed instances that are not SSM managed, scanned every 24 hours, with newly eligible instances detected hourly. Network reachability is assessed every 12 hours.
  • Agent-based prerequisites: SSM Agent installed and running, and an instance profile with the AmazonSSMManagedInstanceCore managed policy. Private subnets need the inspector2-telemetry, s3, ssm and ssmmessages VPC endpoints. AWS recommends upgrading to Enhanced EC2 Scanning (the Amazon Inspector VM Scanner) under Settings, Scan Settings.
  • Agentless eligibility: a supported OS; status Unmanaged EC2 instance, Stale inventory or No inventory; EBS-backed; ext3, ext4 or xfs filesystems; fewer than eight attached volumes totalling 1,200 GB or less. Inspector creates snapshots tagged InspectorScan, reads them through EBS direct APIs and deletes them.
  • Deep inspection: on Linux, programming language packages are found as well as OS packages. Agent-based deep inspection covers default paths plus paths you add; the agentless method reads all available paths, so the same instance can return different findings by method. Windows and macOS instances receive OS package scanning.
  • Exclusions: tag instances with InspectorEc2Exclusion; for agentless scans, tag the KMS key that encrypts the volume. Allow tag access in instance metadata so the SSM plugin honours the tag.

Container images and Lambda

  • Amazon ECR: activating Inspector ECR scanning switches the private registry to enhanced scanning, covering OS and language packages. Continuous scanning triggers on push, on a new relevant CVE and when an image moves from ARCHIVED to ACTIVE. Images stay monitored only within the configured push and re-scan windows, based on the last in use date for recent activations; set the re-scan duration deliberately for long-lived images. Archived images are not scanned and their findings close.
  • Container mapping: Inspector maps images to running ECS (including Fargate) and EKS containers, with in-use counts filterable through ecrImageInUseCount and ecrImageLastInUseAt; data for new images can take up to 36 hours.
  • Lambda: standard scanning covers function and layer dependencies; code scanning (which requires standard scanning) covers custom code. Scans trigger on discovery, deployment, code or dependency updates and new CVEs. Eligible functions use a supported runtime, are $LATEST and were invoked or updated in the last 90 days. Functions encrypted with customer managed keys sit outside both scan types, so cover those with pipeline scanning.

Availability: Inspector is billed per scanned resource; ECR basic scanning is billed through Amazon ECR. Findings flow to AWS Security Hub CSPM and Amazon EventBridge. See the AWS Security Hub Guide and the AWS Security Guide.

Google Cloud: Security Command Center and Artifact Analysis

Compute Engine, GKE and Cloud Run

  • Vulnerability Assessment for Google Cloud: agentless scanning that clones VM disks into a Google-owned project in the source region, mounts the clones in a secure VM and scans them with SCALIBR for OS and language packages and exposed secrets. It covers running Compute Engine VMs, GKE Standard nodes, containers in GKE Standard and Autopilot, and Cloud Run workloads (Preview).
  • Frequency: about every 12 hours on the Premium and Enterprise tiers, weekly on the Standard tier.
  • Eligibility: scanned partitions are VFAT, EXT2, EXT4, XFS and NTFS. CMEK-encrypted disks are scanned when the key is regional in the disk's region, global, or multi-regional in the disk's location, and the Cloud KMS CryptoKey Encrypter/Decrypter role is granted to the SCC organisation service agent and the Compute Engine service agent. Confidential Computing VMs, Confidential GKE Nodes, CSEK disks, GKE nodes on Local SSD ephemeral storage and CMEK disks inside VPC Service Controls perimeters are assessed through agent-based or pipeline scanning instead.
  • Organisation policy: where compute.storageResourceUseRestrictions is enforced, Google's documentation recommends allowlisting the google.com organisation so disk cloning can run.
  • VM Manager: OS Config agent based OS vulnerability reports, surfaced in SCC as OS_VULNERABILITY findings on the Premium tier, give an agent-based view for VMs.

Artifact Registry images

  • Artifact Analysis: enable the Artifact Analysis API and then turn on scanning per repository (it is off by default per repository). New images are scanned once by digest, so retagging does not trigger a rescan.
  • Coverage: OS packages for AlmaLinux, Alpine, CentOS, Chainguard, Debian, Google Distroless, MinimOS, RHEL and UBI, Rocky Linux, SLES, Ubuntu and Wolfi; language packages for Go, Java, Node.js, PHP, Python, Ruby, Rust and .NET. Malicious packages are reported as CRITICAL.
  • Continuous analysis: metadata is refreshed for images pulled in the last 30 days; after that results go stale, and after 90 days stale metadata is archived. Pulling an image refreshes it within up to 24 hours. Use On-Demand Scanning through the gcloud CLI to scan in CI/CD, and Binary Authorization or Cloud Build to block deployment on severity.

Availability: Vulnerability Assessment for Google Cloud runs on the Standard, Premium and Enterprise SCC tiers (automatic on Premium and Enterprise; enabled manually on Standard at organisation, folder or project level). Google states the Enterprise tier is deprecated and moves to Premium on or after 21 May 2027. See the Security Command Center Guide and the Google Cloud Security Guide.

Microsoft Azure: Defender for Cloud and Defender Vulnerability Management

Virtual machines

  • Agent-based: vulnerability assessment for machines with Defender Vulnerability Management, enabled under Environment settings, Defender for Servers, Monitoring coverage, Settings, "Vulnerability assessment for machines". It is on by default when Defender for Servers is enabled.
  • Agentless scanning for machines: Defender for Cloud snapshots OS and data disks, analyses them out of band in the VM's region, deletes the snapshot and passes metadata to MDVM. It reports software inventory, vulnerabilities, EDR configuration, plaintext secrets and, on Defender for Servers Plan 2, malware. It covers Azure VMs and, through connectors, AWS EC2 and GCP Compute Engine instances.
  • Frequency: once every 24 hours on a fixed schedule, and only for running machines; allow up to 24 hours after enabling before troubleshooting.
  • Eligibility: Azure VMs with up to 14 disks; when OS plus data disks exceed 4 TB only the OS disk (under 4 TB) is scanned. Standard VMs and flexible scale sets are covered; VMs with UltraSSD_LRS, PremiumV2_LRS or AKS ephemeral OS disks, UFS, ReFS or ZFS filesystems, RAID members or Databricks VMs need agent-based assessment instead.
  • Customer-managed keys: grant the Microsoft Defender for Cloud Servers Scanner Resource Provider Get, Wrap Key and Unwrap Key (access policies) or Key Vault Crypto Service Encryption User (Azure RBAC).

Containers

  • Registry assessment: Defender for Containers scans ACR, Amazon ECR, Google Artifact Registry and configured external registries once Registry access is enabled, covering OS packages (Linux and Windows) and language packages (Linux). New images are typically scanned within a few hours; a daily rescan covers images pushed or pulled in the last 30 days or running in monitored clusters.
  • Runtime assessment: maps registry findings to running workloads (Registry access plus Kubernetes API access or the Defender sensor), or scans images collected from the runtime regardless of registry (Agentless scanning for machines plus Kubernetes API access or the sensor). Agentless inventory refreshes about every 24 hours; the sensor is near real time. Use registry or pipeline scanning for Windows node containers and AKS ephemeral OS disk nodes.

Availability: agent-based assessment needs Defender for Servers Plan 1 or Plan 2; agentless machine scanning is included in Defender CSPM and Defender for Servers Plan 2; container assessment is part of Defender for Containers, with risk-based prioritisation in Defender CSPM. See the Microsoft Defender for Cloud Guide and the Microsoft Azure Security Guide.

Agent-based or agentless: decision guidance

FactorAgent-basedAgentless (snapshot)
Detection latencyEvent driven: launch, package change, new CVEScheduled: about 12 to 24 hours, or weekly on lower tiers
Deployment effortAgent, instance role and outbound endpoints on every VMOne-time connector or role; no host change
Short-lived instancesScanned at launch if the agent is baked into the imageInstances that live less than a scan cycle may never be assessed
Stopped machinesAssessed when they next runAssessed only when running at scan time (Azure) or when disks are eligible
Encryption and exotic disksReads the live filesystemNeeds key grants and supported filesystems and disk types
Identity footprintPer-host role (for example the SSM instance profile)Central scanner role with snapshot and key permissions
  • Start agentless for discovery: it finds unmanaged and forgotten VMs fast, with no change window.
  • Keep agents where hours matter: internet-facing hosts, regulated systems and golden images, where a new KEV entry should produce a finding the same day.
  • Container images belong in the pipeline: registry scanning is the safety net; scan in CI and gate on policy so vulnerable images never reach the registry. See the Container Image Vulnerability Management Guide.
  • Multi-cloud estates: decide whether each cloud's native scanner feeds a central platform, or one scanner covers several clouds through connectors (Defender for Cloud and SCC both document cross-cloud connectors). Either works; the requirement is one deduplicated queue. See the CNAPP Guide and ASPM vs CNAPP.
  • Small teams: a single cloud with native scanning turned on everywhere, plus a weekly coverage review, beats a third-party tool deployed to half the estate.

Where identity fits: scanner roles and service agents

Agentless scanners work because a provider-owned identity is allowed to snapshot or clone your disks and, for encrypted disks, to use your keys. These are powerful non-human identities. Inspector uses a service-linked role and creates InspectorScan snapshots; Defender for Cloud uses the VM scanner operator role on Azure, a VmScanner role deployed by CloudFormation on AWS and a custom MDCAgentlessScanningRole on Google Cloud; SCC uses its organisation or project service agent plus the Compute Engine service agent for CMEK.

  • Inventory them: record each scanner role, service agent and key grant with an owner and the reason it exists.
  • Scope key grants: grant decrypt only on the keys for in-scope disks, and review the grants when keys rotate.
  • Watch the audit trail: snapshot creation and deletion by scanner identities is expected; snapshot sharing, copying across accounts or long-lived scanner snapshots are not. On AWS, Defender suspends scans when 250 or more of its tagged snapshots accumulate, so alert on snapshot clean-up failures.
  • Agent roles too: SSM instance profiles and OS Config service accounts grant management-plane reach to every host; keep them least privilege. See the Non-Human Identity (NHI) Guide.

Implementing native scanning across clouds

  1. Inventory first: export the compute, registry and serverless inventory per account, project and subscription (CM-8). This is the denominator for every coverage metric.
  2. Enable at the organisation level: use the delegated administrator in Inspector, organisation activation in SCC and management group or connector scope in Defender for Cloud, so new accounts inherit scanning.
  3. Pilot one ring: turn on agentless scanning in a non-production account or project, confirm snapshots are created and cleaned up, and check findings against a known vulnerable test image.
  4. Grant encryption access: apply the documented key grants for CMK and CMEK disks; re-run coverage and confirm encrypted VMs move from unscanned to scanned.
  5. Close agent gaps: bake agents (for example SSM Agent on AWS and the OS Config agent on Google Cloud, and the agent-based assessment option in Defender for Servers on Azure) into golden images for hosts that need continuous results.
  6. Set container windows: configure ECR re-scan duration, Artifact Registry per-repository scanning and Defender registry access; schedule pulls or rebuilds for long-lived images so they stay inside continuous analysis windows.
  7. Route and deduplicate: send findings to Security Hub, SCC and Defender for Cloud, then to one ticketing queue keyed on CVE, asset and image digest.
  8. Roll back safely: agentless scanning can be switched off per scope without host changes; record exclusions with owner, reason and expiry rather than disabling scanning wholesale.

Detection and measurement

  • Coverage ratio: scanned workloads divided by inventory, per workload type and per cloud. Target 95 percent or more for VMs and running images.
  • Unscanned by reason: counts of workloads excluded by filesystem, disk limit, key access, image age or tag. Each reason needs an owner and a compensating scan.
  • Scan freshness: share of assets with a "last scanned" time within one scan cycle (Inspector ListCoverage, SCC finding timestamps, Defender assessment time).
  • Time to remediate: median days from first finding to fix for critical and KEV-listed CVEs, per cloud.
  • Running vulnerable images: number of critical findings on images actually in use, from container mapping or runtime assessment, not total registry findings.
  • Scanner health: snapshot clean-up failures and scanner permission errors, alerted on rather than discovered in quarterly reviews.

See Vulnerability Management Metrics for definitions and the Risk-Based Vulnerability Management Guide for prioritisation.

Framework mapping

FrameworkReferenceHow native scanning supports it
NIST CSF 2.0ID.RA-01Vulnerabilities in cloud workloads identified, validated and recorded
NIST CSF 2.0ID.AM-01, PR.PS-02Inventory reconciliation; software maintained and replaced by risk
NIST SP 800-53 Rev 5RA-5, SI-2, CM-8Vulnerability monitoring and scanning; flaw remediation; component inventory
CIS Controls v8.1Control 7 (Continuous Vulnerability Management)Automated scanning of internal and cloud assets and remediation tracking
ISO/IEC 27001:2022Annex A 8.8Management of technical vulnerabilities
MITRE D3FENDD3-AVE Asset Vulnerability EnumerationEnumerating vulnerabilities per asset; reduces exposure to T1190 Exploit Public-Facing Application

Common failure patterns

  • Treating a console toggle as coverage: VM scanning is on, but registry scanning is off per repository, or Lambda functions with customer managed keys sit outside scope, and nobody checks the coverage list.
  • Encrypted disks silently skipped: CMK or CMEK disks stay unscanned because the documented key grant was never applied, often on the most sensitive systems.
  • Old images aging out: long-running images fall outside push, pull or re-scan windows and stop receiving new CVE matches while still running in production.
  • Short-lived instances never assessed: autoscaled hosts that live less than a scan cycle are only covered if the golden image carries an agent or is scanned in the pipeline.
  • Organisation policy blocks: storage resource or key restrictions stop disk cloning or snapshots, and the failure only shows as missing findings.
  • Three clouds, three queues: findings stay in each provider console, the same base image CVE is ticketed three times, and owners ignore the duplicates.

Practitioner checklist

  • Enable native scanning at organisation scope in every cloud so new accounts, projects and subscriptions inherit it.
  • Reconcile the scanner coverage list against the asset inventory weekly, per workload type.
  • Apply the documented key grants for CMK and CMEK disks and confirm encrypted VMs appear as scanned.
  • Bake agents into golden images for internet-facing, regulated and short-lived workloads.
  • Turn on per-repository registry scanning and set re-scan windows that match how long images run.
  • Scan images in CI and gate deployment, keeping registry scanning as the backstop.
  • Inventory scanner roles, service agents and key grants as non-human identities with owners.
  • Alert on snapshot clean-up failures and scanner permission errors.
  • Send all findings into one deduplicated queue keyed on CVE, asset and image digest.
  • Record every exclusion with owner, reason, compensating scan and expiry.

Standards and references

Related NHI Mgmt Group resources: Vulnerability Scanning Guide · CNAPP Guide · ASPM vs CNAPP · Container Image Vulnerability Management Guide

PDF for subscribers

Take this guide with you

Every section, table and checklist in one PDF, with links to the sources and related guides. Subscribe to our weekly NHI & AI Identity Journal and we’ll email you a download link.

Free weekly newsletter

Subscribe to the NHI & AI Identity Journal

The latest on NHI and Agentic AI security – articles, research, breaches, news and events every week.

Bonus 33% off our NHI Course when you subscribe.

NHIMG Editorial Note
Researched and written by the NHI Mgmt Group editorial team using AI tools, under the editorial direction of Lalit Choda. Facts are drawn from the sources listed on this page and checked by our automated quality process. Last updated 14 November 2026. Spotted an error? Use our contact page.
NHI Mgmt Group, the independent authority on Non-Human Identity, IAM and Agentic AI security. nhimg.org

PDF Downloads for Newsletter Subscribers

Get the guide

Enter your details and we’ll email you a download link.

The full guide is free to read online. The PDF is a thank you for subscribing.

Fill in all fields marked * and tick the PDF box to continue.