Join our Newsletter — 33% off our NHI Course
Home› Guides› Docker Security Checklist: Host, Daemon, Images, Runtime and…
Template Cloud Security

Docker Security Checklist: Host, Daemon, Images, Runtime and Registry Checks

PDF for newsletter subscribers · free to read online

← All guides
By NHI Mgmt Group, under the editorial direction of Lalit Choda Updated 18 November 2026 13 min read
On this page

This Docker security checklist is a printable, tick-box review of one Docker host or fleet, covering the host and daemon, the API socket, images and builds, runtime flags, registries, secrets and logging, with an evidence column for every check. It is designed for platform engineers, security reviewers and auditors who need a repeatable record rather than a tutorial. The reasoning behind each control is covered in the Docker and Docker Compose Security Guide; this page is the list you work through and sign.

Checks are based on Docker's public Engine security documentation as of November 2026, plus NHI Mgmt Group recommendations where marked. Docker features change between releases, so confirm settings against Docker's current documentation for the version you run.

Key takeaways

  • Treat access to the Docker daemon as root access to the host. Docker's own documentation says the docker group grants root-level privileges.
  • Never expose the API without mutual TLS or SSH. Docker states that exposing the daemon API over HTTP without TLS is not permitted, and the daemon fails at startup in that configuration.
  • Rootless mode is the strongest single host control, because the daemon and containers run without root inside a user namespace.
  • Most runtime risk comes from a handful of flags: --privileged, added capabilities, host mounts, host networking and running as root.
  • Secrets and registry credentials are non-human identities. Inventory them, scope them and rotate them like any other credential.
  • A checklist is only useful with evidence. Record the command, output or screenshot for every tick.

How to use this template

  1. Copy the tables into your review document or ticket. Complete one copy per host class (for example build hosts, production hosts, developer workstations) rather than one per machine.
  2. For each check, mark Pass, Fail, N/A or Exception, and record the evidence. An exception needs an owner, a reason and an expiry date.
  3. Rate each fail as High, Medium or Low using the suggested rating in the table, adjusted for exposure. An internet-facing host raises every rating by one level.
  4. Carry fails into the findings log in section 9 and agree remediation dates.
  5. Repeat the review at least every six months, after a major Docker Engine upgrade, and after any incident involving a container host. These intervals are NHI Mgmt Group recommendations.

Kubernetes clusters need their own review; use the Kubernetes Security Checklist for orchestration controls and keep this list for the Docker Engine layer.

1. Review record

FieldValue
Host class or fleet name
Docker Engine version and install method
Host operating system and kernel version
Rootful or rootless
Workload owner
Reviewer and date
Scope exclusions
Previous review reference

2. Host and daemon

#CheckHow to verifyRating if failedResult and evidence
2.1The host runs only container workloads and the tools needed to manage themReview installed packages and listening servicesMedium
2.2Docker Engine and the host kernel are on supported, patched versionsdocker version, uname -r, compare with vendor support policyHigh
2.3Membership of the docker group is limited to named administrators and reviewed, because the group grants root-level privilegesgetent group docker; compare with approved listHigh
2.4Rootless mode is used where the workload allows it, or the reason for rootful mode is recordedCheck the daemon runs under a non-root user; confirm the system-wide daemon is disabledMedium
2.5If rootless mode is used, the system-wide rootful daemon is not also runningsystemctl status docker for the system serviceMedium
2.6Where rootful mode is required, user namespace remapping is considered and the decision recordedReview daemon.json for userns-remapLow
2.7The daemon configuration file is owned by root and not writable by other usersls -l /etc/docker/daemon.jsonMedium
2.8A mandatory access control system (AppArmor or SELinux) is enabled on the hostdocker info security options; aa-status or getenforceMedium
2.9Docker data directories sit on a dedicated volume or partition so a full disk cannot take down the hostdf -h for the Docker root directoryLow

Docker's rootless documentation explains that in user namespace remapping mode the daemon itself still runs with root privileges, while in rootless mode both the daemon and containers run without root. That is why 2.4 is preferred over 2.6 where workloads allow it. Rootless mode also needs newuidmap and newgidmap and at least 65,536 subordinate IDs in /etc/subuid and /etc/subgid.

3. Daemon socket and remote API

#CheckHow to verifyRating if failedResult and evidence
3.1The daemon listens only on the local Unix socket unless remote access is approvedInspect daemon.json hosts and service unit flags; ss -lntpHigh
3.2Where remote access is needed, it uses SSH or mutual TLS with tlsverify and a trusted CAReview daemon flags and certificate configurationHigh
3.3TLS server certificates match the hostnames and IPs clients use, and client certificates are issued per person or per systemInspect certificates; compare with issuance recordsMedium
3.4Client certificates and SSH keys used for Docker access are inventoried with owners and expiryCredential inventory recordMedium
3.5The Docker socket is not mounted into containers, or each exception is documented and restrictedSearch running containers for /var/run/docker.sock mountsHigh
3.6Network firewall rules block the Docker API ports from untrusted networksFirewall or security group reviewHigh
3.7An authorisation plugin or access proxy is considered for shared hosts that need finer controlDesign recordLow

Docker notes that by default it runs through a non-networked Unix socket and that remote access should use SSH or a TLS socket. Mounting the socket into a container hands that container full control of the daemon, which is why 3.5 is rated High. Teams often find socket mounts in monitoring agents, CI runners and reverse proxies.

4. Images and builds

#CheckHow to verifyRating if failedResult and evidence
4.1Base images come from an approved list of trusted publishers or an internal mirrorReview FROM lines against the approved listMedium
4.2Images are referenced by digest or by a pinned version tag, not latestSearch Dockerfiles and Compose filesMedium
4.3Images are scanned for known vulnerabilities in the build pipeline and before deployment, with a blocking policy for critical findingsPipeline configuration and recent resultsHigh
4.4Images are scanned for hard-coded secrets, and any finding triggers rotation, not just a rebuildSecret scan results; rotation ticketsHigh
4.5Dockerfiles set a non-root USERReview Dockerfiles; docker inspect Config.UserMedium
4.6Images are minimal: no compilers, shells or package managers in production images unless neededImage contents reviewLow
4.7Build secrets use build secret mounts, not ARG, ENV or copied filesDockerfile review; docker historyHigh
4.8Images are signed and signatures are verified before deploymentSigning and verification configurationMedium
4.9A software bill of materials is produced and stored for each released imageSBOM artefacts in the registry or pipelineLow

Secrets baked into image layers are one of the most common container findings, because deleting a file in a later layer does not remove it from earlier ones. See Secrets in Container Images and Build Artefacts Guide for the clean-up process.

5. Container runtime flags

#CheckHow to verifyRating if failedResult and evidence
5.1No production container runs with --privileged without a documented exceptiondocker inspect --format '{{.HostConfig.Privileged}}' for each containerHigh
5.2Containers drop all capabilities and add back only those requiredInspect CapDrop and CapAddMedium
5.3No container adds SYS_ADMIN or other broad capabilities without an exceptionInspect CapAddHigh
5.4Containers run as a non-root user (--user or image USER)Inspect Config.User; docker exec idMedium
5.5The default seccomp profile is in place; no container runs with seccomp unconfinedInspect SecurityOptHigh
5.6The no-new-privileges security option is set (NHI Mgmt Group recommendation)Inspect SecurityOptMedium
5.7Root filesystems are read-only where the application allows, with writable paths as named volumes or tmpfs (NHI Mgmt Group recommendation)Inspect ReadonlyRootfsLow
5.8Host paths are not bind-mounted, especially /, /etc, /proc or /var/run; any needed mount is read-only and narrowInspect MountsHigh
5.9Host network, PID and IPC namespaces are not shared without an exceptionInspect NetworkMode, PidMode, IpcModeHigh
5.10Memory, CPU and process limits are set for each containerInspect Memory, NanoCpus, PidsLimitMedium
5.11--oom-kill-disable is never used without a memory limitInspect OomKillDisable and MemoryMedium
5.12Published ports bind to specific interfaces, not all addresses, unless the service is meant to be publicInspect PortBindingsMedium
5.13Devices are not passed through except where required and documentedInspect DevicesMedium

Docker's run reference states that --privileged gives all capabilities to the container, along with access to all host devices and loosened AppArmor or SELinux confinement, and that the default user in a container is root. The seccomp documentation describes the default profile as an allowlist that disables around 44 system calls out of more than 300 and advises against changing it. Docker also warns that disabling the out-of-memory killer without a memory limit can exhaust host memory. Container escape maps to MITRE ATT&CK T1611 Escape to Host, and most escape paths start with one of the flags in this table.

6. Registry and pulls

#CheckHow to verifyRating if failedResult and evidence
6.1Hosts pull only from approved registriesDaemon and pipeline configuration; network egress rulesMedium
6.2Registry connections use TLS; no insecure registries are configureddocker info insecure registries; daemon.jsonHigh
6.3Pull credentials are read-only and scoped to the repositories each host needsRegistry token or role configurationMedium
6.4Push credentials exist only in the build pipeline, not on runtime hosts or developer laptopsCredential inventory; search ~/.docker/config.json on hostsHigh
6.5Registry access tokens have owners, expiry dates and a rotation recordToken inventoryMedium
6.6Credential helpers are used instead of plain-text auth entries in config.jsonInspect config.json for auths entries with stored valuesMedium
6.7Private repositories are private; public repositories are reviewed for accidental secretsRegistry visibility settings; secret scanHigh

Registry tokens are a frequent source of leaks. The Docker Hub Tokens Security Guide and Container Registry Security Guide cover scoping and rotation in depth.

7. Secrets and identity

#CheckHow to verifyRating if failedResult and evidence
7.1Application secrets are not passed as plain environment variables in Compose files or run commands checked into source controlSearch repositories and docker inspect Config.EnvHigh
7.2Secrets are delivered at runtime from a secrets manager or as mounted secret files with restricted permissionsDeployment configurationMedium
7.3Each container workload has its own credentials, not a shared credential across servicesCredential inventoryMedium
7.4Cloud credentials used by containers are short-lived (workload identity or instance roles), not long-lived keysCredential type reviewHigh
7.5Containers cannot reach the cloud instance metadata service unless they need itNetwork policy or host firewall testMedium
7.6Every secret found in this review has an owner and a rotation dateFindings logMedium

Each container that calls an API or a database is a non-human identity. A compromised container inherits whatever its credentials can do, which is why scope and lifetime matter more than where the secret is stored.

8. Logging and monitoring

#CheckHow to verifyRating if failedResult and evidence
8.1Daemon logs are forwarded to central loggingLog pipeline configuration; sample eventsMedium
8.2Container logs use a driver with size limits or central forwarding so disks do not filldocker info logging driver; log-optsLow
8.3Host audit rules cover the Docker binary, daemon configuration, socket and data directoryAudit rule listingMedium
8.4Alerts exist for privileged container starts, socket mounts and new containers from unapproved imagesDetection rule listMedium
8.5Alerts exist for exec into production containersDetection rule listLow
8.6A runtime security tool or host EDR covers container processesAgent deployment recordMedium

Attackers who reach the API often start new containers, which maps to MITRE ATT&CK T1610 Deploy Container. Detection for this needs daemon events, not just container logs. The Container Runtime Security Guide covers runtime monitoring options.

9. Findings and sign-off

Check #FindingRatingOwnerRemediationDue dateStatus

Suggested remediation windows (NHI Mgmt Group recommendation): High within 14 days, Medium within 60 days, Low at the next planned change. Sign-off requires the reviewer, the workload owner and the platform lead to confirm the record.

RoleNameDate
Reviewer
Workload owner
Platform lead

Common failure patterns

  • Docker group as a convenience. Developers are added to the docker group to avoid typing sudo, which hands them root on the host.
  • Socket mounted for monitoring. A dashboard or CI runner gets /var/run/docker.sock, and a compromise of that tool becomes control of every container.
  • Privileged as a fix. --privileged is added to make a failing container work and never removed.
  • Rootless installed alongside rootful. Rootless mode is set up but the system daemon keeps running, so the host is still rootful.
  • Secrets in image history. An ARG or copied file leaks a token into a layer, and the image is pushed to a shared registry.
  • Push tokens everywhere. The same registry token with push rights sits on build hosts, runtime hosts and laptops.
  • Ticks without evidence. The checklist is marked complete from memory, and the next incident shows half the passes were wrong.

Practitioner checklist

  • List every member of the docker group on each host and remove anyone who does not need root.
  • Confirm the daemon listens only on the Unix socket, or on SSH or mutual TLS where remote access is approved.
  • Search all running containers for Docker socket mounts and privileged mode, and raise exceptions or fixes.
  • Move suitable hosts to rootless mode and disable the system daemon on them.
  • Enforce non-root users, dropped capabilities, the default seccomp profile and resource limits in deployment templates.
  • Scan images for vulnerabilities and secrets in the pipeline, and rotate any secret found.
  • Restrict push credentials to the build pipeline and give runtime hosts read-only pull tokens.
  • Forward daemon events to central logging and alert on privileged starts and socket mounts.
  • Record evidence for every check and repeat the review every six months.

Standards and references

Related NHI Mgmt Group resources: Docker and Docker Compose Security Guide · Container Runtime Security Guide · Docker Hub Tokens Security Guide · Kubernetes Security Checklist

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.

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 both boxes to continue.