Are untagged container images safe to delete from a registry?

Usually, but check two things first. Anything that pulls by digest will break if the untagged manifest disappears, and untagged manifests can be children of a multi-architecture image that still has a tag. Confirm both, then delete after a grace period.

What untagged means

A tag is a movable label on a manifest. When you push a new image to an existing tag, the old manifest loses that tag and becomes untagged. It is still stored and still reachable by its digest. Registries therefore offer cleanup for untagged manifests, since they pile up fast on busy repositories.

Risk one: pulling by digest

Pinning by digest is a good practice, and many deployments do it. Azure Container Registry warns that if you have systems that pull images by manifest digest, you should not set a retention policy for untagged manifests, because deleting untagged images stops those systems from pulling. A digest pinned deployment may reference a manifest that currently has no tag at all.

So before cleanup, search deployment files, Helm values and admission policies for digest references, and compare against untagged manifests. See pinning base images by digest for why teams do this.

Risk two: multi-architecture children

A multi-arch image is an index that points to one manifest per platform. Those per platform manifests can look untagged, yet they are referenced. Amazon ECR documents that an image referenced by a manifest list cannot be expired without the manifest list being deleted first. Artifact Registry states that cleanup policies for Docker images do not delete images referenced by a parent manifest, and related images go when the parent is deleted. Good registries protect against this, but verify your own.

Risk three: attached artifacts

Signatures, SBOMs and attestations can be stored as separate manifests that point at the image. See does deleting an image delete its SBOM and signature.

A safe procedure

1. List untagged manifests and their ages.
2. Search your deployment sources for digest references.
3. Start with a conservative age so recent pushes survive.
4. Use the preview or dry run if your registry has one.
5. Enable the rule and watch for pull failures in the following days.

Details that differ by registry

Azure documents that its untagged manifest retention applies only to untagged manifests with timestamps after the policy is enabled, so older ones are not affected, and that OCI media type manifests are not supported by that feature while Docker manifests are. It points to the purge command for OCI manifests. In ECR, only one rule that selects a given storage class may select untagged images. Read the page for the registry you use.

Where Safeguard fits

Safeguard's asset discovery and SBOM tooling help you see which images and digests are in use before you remove anything. A free scan is at gold.safeguard.sh/scan.

Record the decision

Write the grace period and the checks you ran in the repository's runbook. When a pull failure appears later, the first question will be whether a cleanup rule caused it, and a dated note answers that in a minute. Also tell the teams that deploy from the repository, since they are the ones who will see an image pull error first.

Quick check

Sources

• Azure Container Registry retention for untagged manifests
• Amazon ECR lifecycle policies
• Artifact Registry cleanup policies

Last reviewed: 19 October 2026.