Why this matters
The Cloud console will create a VM from an old image and never tell you the packages on it are out of date. Vulnerabilities in Google Cloud are not surfaced by one product. Security Command Center covers posture and web app scanning, Artifact Analysis covers container images, and the OS Config agent feeds package inventory into VM Manager. Turn on one and you see a slice of the truth.
Partial views are where incidents come from. A base image from last spring still runs in a side project. An internal tool got a public load balancer during a demo and kept it. Scanning finds those before a customer's security team does.
The hard part is the output. A first scan of a real estate returns more findings than anyone can fix in a sprint, and sorting by severity alone misleads. A critical CVE in a package nothing loads matters less than a medium finding on an internet-facing service. Pair every finding with exposure and an owner.
Before you start
Pick the scope first. Security Command Center turns on at the organization, folder or project level, and settings roll down. Enable it at the organization so new projects are covered without a ticket, and write down any documented exceptions. An unexplained gap is what auditors ask about.
Decide your tier. The free Standard tier is mostly asset discovery. Web Security Scanner, VM Threat Detection, Event Threat Detection and the fuller set of Security Health Analytics detectors sit in the paid Premium or Enterprise tier. Names and contents change, so open Security Command Center, go to Settings, and read what is active before you promise anyone a report.
Check the prerequisites. VM Manager needs the OS Config agent running on each instance and the OS Config API enabled. Container scanning needs images in Artifact Registry with scanning on for the repository. Web scanning needs a target reachable from outside, plus an authentication profile for pages behind a login. Test in a sandbox project, and decide where findings go and who owns them before you switch anything on.
How to do it
Work from the outside in. Enable the scanners, confirm coverage, then build the plumbing that makes findings useful: exports, owners, mute rules for accepted risk. A scan you cannot route to a person is a report, not a process.
Container images give the quickest return. Turn on scan on push for every Artifact Registry repository and schedule a rescan for images already deployed. The digest does not change, but the CVE feed does, so an image that was clean in March can be critical in June.
For VMs, the goal is installed packages compared against known vulnerabilities, per instance. VM Manager gives you that, and patch compliance if you run patch jobs. Test on a canary group first so you learn how noisy your golden image is. Web scanning is the last piece and needs the most setup: public URLs, an auth profile, and the scanner's source ranges allowed at your firewall or load balancer.
- Confirm your Security Command Center tier. Open Security Command Center in the console and go to Settings. Note which services are active and which finding sources you actually have. Write that list down so the next step has a target.
- Turn on Security Command Center at the organization. Activate services at the organization node, not just on individual projects. Check that folder and project pages show inherited settings. New projects should appear in Asset Inventory without extra work.
- Create a read-only scanner service account. Make a dedicated service account for automation and give it roles like Security Center Finding Viewer and Artifact Registry Reader. Do not reuse an owner or editor account. Keep its key or workload identity binding in your secret manager.
- Enable container scanning on every repository. In Artifact Registry, turn on vulnerability scanning for each repository, including ones your CI pipeline creates. Push a test image and confirm a scan runs. Look at the scan results for that image digest.
- Run an on-demand scan on deployed images. Use the on-demand scanning API or the gcloud artifacts docker images scan command against images you already run in production. Compare the result with the on-push scan. Schedule this as a recurring job rather than a one-off.
- Enable VM Manager and read the reports. Confirm the OS Config API is on and the OS Config agent is running on your instances. Open VM Manager vulnerability reports for a project and sort by installed package. Cross-check a few packages against the machine itself.
- Set up Web Security Scanner for public URLs. Add the endpoints you care about, choose a schedule, and create an authentication profile for pages behind a login. Allowlist the scanner's published source ranges at your load balancer or firewall. Keep the request rate low on production targets.
- Export findings and mute accepted risk. Create a Pub/Sub or BigQuery export for active findings. Add mute rules for findings you have formally accepted, such as a CVE in a package you never load. Route everything else to the team that owns the resource.
- Fix one finding and rescan it. Pick a high finding with a clear fix, patch the image or the VM, then rescan. Confirm the finding moves to inactive. That proves the loop from detection to verified fix, which is what the next audit asks about.
How to check it worked
You know it works when a deliberately bad image produces a finding with nobody clicking a button. Push an image with an old base, wait for the scan, and look for an ACTIVE finding in Security Command Center. If nothing shows up, the repository is not scanning, the API is off, or findings have not synced at your tier.
Check VM coverage by comparing the Compute Engine instance list with the VM Manager list. Missing hosts either lack a running OS Config agent or sit in a project with the API disabled. On one host, run the package query locally and compare a few versions with the report.
Check the export path with a real finding. Read the Pub/Sub payload and confirm your downstream job parses the finding name, category, severity and resource. Then check gaps on purpose: in Asset Inventory, look for projects with assets but no findings. A quiet project is usually an unscanned one.
Mistakes to avoid
Enabling at project level and stopping there. Projects created later by a platform team stay dark until someone remembers. Organization level enablement with written exceptions avoids this.
Scanning only at build time. Images built before scanning existed, or running for a year, are the ones that bite. Add a scheduled rescan and a way to rebuild from a patched base.
Treating severity as priority. Weigh exposure and reachability first: public IP, internet-facing load balancer, broad service account. Then sort by severity.
Letting noise win. Findings with no owner get ignored within weeks. Assign each finding source to a team, mute accepted risks with a comment and an expiry date, and review that list.
Relying on one tool. Google's built-in scanners do not read your Terraform, your application code or your source dependencies, so pair them with a code scanner. If you want an outside read-only view of identity, network exposure, encryption and logging across up to five cloud accounts, with findings ranked by risk and a 60-minute readout, the Cloud security review package from Cloudverg is $6,500/month, fixed scope, delivered in 2 to 3 weeks.
Frequently asked questions
Does Google Cloud scan for vulnerabilities by default?
Mostly no. Artifact Analysis scans images on push only when you enable scanning on the repository, and VM package vulnerabilities appear only after the OS Config agent reports inventory to VM Manager. Web application scanning sits in a paid Security Command Center tier. Treat scanning as something you configure, not something you inherit.
What is the difference between Security Command Center and Artifact Analysis?
Security Command Center is the aggregation and posture layer. It collects findings from Google services and shows them per organization, folder or project. Artifact Analysis is the container scanner that inspects image layers and installed packages. Container findings normally show up in Security Command Center once an image is deployed, not only in the registry.
How often should I rescan container images?
Scan on every push, and rescan images already running on a schedule. CVE data changes daily while your image digest does not, so a weekly rescan of deployed images catches problems that arrived after the build. Rebuild from a patched base instead of patching inside a running container.