Join our Newsletter — 33% off our NHI Course
Home› Guides› OWASP CI/CD Top 10 CICD-SEC-10 Insufficient Logging and…
Guide Software & AI Supply Chain

OWASP CI/CD Top 10 CICD-SEC-10 Insufficient Logging and Visibility Guide: Logging and Monitoring the Pipeline

PDF for newsletter subscribers · free to read online

← All guides
By NHI Mgmt Group, edited by Lalit Choda Updated 26 October 2026 14 min read
On this page

Most organisations log their servers, firewalls and business applications in depth, yet cannot say who changed a pipeline, which job read a secret or which identity published last night's artifact. CICD-SEC-10 Insufficient Logging and Visibility is the tenth risk in the OWASP Top 10 CI/CD Security Risks, and it decides whether the other nine are caught at all. This guide explains what OWASP asks for, which audit and application logs each CI/CD system can produce (and which are off by default), how to bring human and machine identity activity into one view, and how to show an auditor the control works. It builds on the Top 10 CI/CD Security Risks and the CI/CD Security Guide, and is written for detection engineers, platform teams and the security architects who own software delivery.

Key takeaways

  • Two kinds of log, equal weight. OWASP asks for both system audit logs (user access, user creation, permission changes) and applicative logs (pushes, build executions, artifact uploads); most pipeline attacks only show up when the two are joined.
  • Logging is not on by default. OWASP notes that relevant logs are not enabled by default; Azure DevOps auditing, GitHub source IP disclosure and GitHub API request streaming are concrete examples that need a deliberate switch.
  • Inventory comes first. You cannot monitor a self-hosted Jenkins controller, container registry or Kubernetes cluster you do not know about, so the first recommendation is a full inventory of every CI/CD system instance.
  • Machine identities do most of the work. Pipelines act through tokens, deploy keys, bots and service accounts, so visibility must cover programmatic access as thoroughly as human sign-ins.
  • Retention defaults are short. Native retention is 180 days for a GitHub organisation audit log and 90 days for Azure DevOps, so stream to a central store early or lose the evidence a slow investigation needs.

What it is and who it applies to

OWASP defines the risk as insufficient logging and visibility that lets an adversary act inside the CI/CD environment without being detected at any stage of the kill chain, and that leaves the organisation unable to reconstruct the attacker's techniques afterwards. The project's description makes a pointed observation: workstations, servers, network devices and key business applications are usually covered by logging programmes, but engineering systems and processes often are not.

The risk applies to any organisation that builds or deploys software through automation, including those that only consume SaaS CI/CD. The scope OWASP lists is broad: source control management (SCM), CI, artifact repositories, package management, container registries, CD and orchestration engines such as Kubernetes. Self-managed systems, Jenkins being OWASP's own example, need special attention because every instance has to be found and configured separately.

The OWASP text is short and closes by stating that logging is relevant to every incident regardless of which risk was exploited. That is the practical reason this entry matters: CICD-SEC-10 is the detective control behind CICD-SEC-1 Insufficient Flow Control Mechanisms, CICD-SEC-4 Poisoned Pipeline Execution and CICD-SEC-6 Insufficient Credential Hygiene.

Structure and key requirements

OWASP gives four recommendations. The table restates each one with what it means in practice.

OWASP recommendationWhat it requiresPractical output
Mapping the environmentInventory every system involved in CI/CD, including every instance of self-managed systemsA register of SCM organisations, CI controllers and runners, registries, CD tools and clusters, each with an owner and a log status
Identifying and enabling log sourcesEnable audit and applicative logs, covering human and programmatic access by every method allowedA log source matrix per system: what is enabled, at what level, where it goes
Shipping logs to a centralised locationAggregate in a SIEM or equivalent to correlate across systemsStreams or collectors for each source, with parsing and time synchronisation
Creating alertsDetect anomalies in each system and in the code shipping process as a wholePer-system rules plus cross-system rules that need knowledge of how builds and deployments normally flow

Log sources by system

The second table lists what the main platforms record and the settings that matter. Confirm the details for your edition and plan before relying on them.

SystemAudit and application eventsSettings and limits to know
GitHub organisationAudit log categories such as repo, org, protected_branch, workflows, hook, oauth_application, personal_access_token, packages180 days of events; export to JSON or CSV; source IPs hidden until "Enable source IP disclosure" is set; API access to the audit log needs GitHub Enterprise Cloud
GitHub enterprise streamingAudit events and Git events across the enterpriseDestinations include Amazon S3, Azure Blob Storage, Azure Event Hubs, Datadog, Google Cloud Storage and Splunk; "Enable API Request Events" is a separate opt-in and covers security-relevant endpoints only; a paused stream buffers for seven days
GitLabAudit events for a top-level group, its subgroups and projectsAudit event streaming (Ultimate) to an HTTP endpoint, Google Cloud Logging or AWS S3; verify the X-Gitlab-Event-Streaming-Token header; deduplicate on the event id
JenkinsAudit Trail plugin: job create, configure and delete, build starts and triggers, credential usage, Groovy script use, logins, failed logins and logoutsFile, syslog, console or Elastic Search logger; set X-Forwarded-For on any reverse proxy or the client IP is lost
Azure DevOpsPermission changes, deleted resources, branch policy changes, log access and downloadsOff by default: enable "Log Audit Events" under Organization settings, Policies; events kept 90 days; audit streaming to Splunk or a SIEM
AWS CodePipelineAPI calls such as CreatePipeline, UpdatePipeline, GetPipelineExecution recorded by CloudTrail with caller and source IPCreate a trail to S3 for an ongoing record; Event history alone is for recent events
KubernetesAPI server audit events, staged per requestAudit policy levels None, Metadata, Request, RequestResponse; log backend (--audit-log-path, --audit-log-maxage) or webhook backend; managed clusters expose this through the provider's logging

Identity and NHI requirements

OWASP stresses that many CI/CD attack vectors use programmatic access, so visibility must be built around human and machine access alike. In a typical pipeline, most actions are taken by non-human identities: CI service accounts, workflow tokens, deploy keys, OAuth apps, bots, runner identities and cloud roles assumed through federation. If the logs only attribute actions to a username, a stolen token looks exactly like normal automation.

  • Attribute every action to an identity type: tag events as human, bot, OAuth app, personal access token, deploy key or workload identity. GitHub's audit log has separate oauth_application and personal_access_token categories for this reason. See the CI/CD Pipeline Identity Security Guide.
  • Log credential use, not only credential changes: the Jenkins Audit Trail plugin can record credential usage through the Credentials plugin, which answers "which job read this secret". Pair it with secrets manager access logs.
  • Follow the token into the cloud: when a pipeline assumes a cloud role, the cloud audit trail (for example AWS CloudTrail AssumeRoleWithWebIdentity) is where the deployment's real permissions are exercised. Join it to the run ID that requested it. See the Workload Identity Federation Guide.
  • Watch token lifecycle events: creation of personal access tokens, deploy keys and new OAuth app grants are persistence signals. See the OWASP NHI Top 10 NHI2 Secret Leakage Guide.
  • Separate human use of machine credentials: a CI service account signing in interactively, or from a laptop rather than a runner network, is a high-value alert. See the OWASP NHI Top 10 NHI10 Human Use of NHI Guide.

The NSA and CISA information sheet on defending CI/CD environments states the same goal from the audit angle: the log should show who committed, reviewed and deployed what, when and where, and administrative accounts should be audited regularly.

Implementation steps

  1. Build the inventory. List SCM organisations and groups, CI controllers and runner pools, artifact and package repositories, container registries, CD tools and clusters. Include self-hosted instances found through network scans and cloud asset inventories, not only those the platform team declares.
  2. Write the log source matrix. For each system record the audit log, the applicative log, the current state (on, off, partial), retention, and the identity types it can attribute. Mark gaps such as Azure DevOps auditing being off, GitHub source IPs hidden, or Jenkins instances without the Audit Trail plugin.
  3. Enable sources in a pilot. Turn on audit logging and streaming for one business unit first. Check volume, parsing and cost. For Kubernetes, start with a Metadata level baseline and raise to Request or RequestResponse only for sensitive resources, because full bodies can capture secret values.
  4. Centralise and normalise. Stream to the SIEM or a security data lake, deduplicate (GitHub and GitLab both state delivery can repeat an event), synchronise time and map fields to a common schema: actor, identity type, repository, pipeline, run ID, commit SHA, artifact digest, source IP. See the Security Data Pipeline Guide.
  5. Protect the logs. Write to storage the pipeline identities cannot modify, restrict who can pause or delete streams, and alert when a stream is paused or an audit setting is turned off. Disabling logging is itself an ATT&CK technique, T1685 Disable or Modify Tools.
  6. Write per-system detections. Start with high-signal events: branch protection or ruleset removal, new admin or owner, new deploy key or webhook, workflow file changes on default branches, self-hosted runner registration, audit stream changes.
  7. Write process detections. Correlate across systems: an artifact in the registry with no matching CI run, a production deployment from a commit that never passed review, a run that read secrets it has never read before, or a pipeline identity calling cloud APIs outside its normal set.
  8. Test and tune. Run purple-team exercises in a sandbox organisation (for example a workflow change by a test account) and confirm each alert fires and reaches the owning team. Keep new rules in alert-only mode for two to four weeks before paging on them.
  9. Roll back safely. If a source floods the SIEM, reduce its level or filter at the collector rather than turning it off, and record the change. Never drop audit events entirely to save cost.

Detection starting points

SignalSourceWhy it matters
New or changed file under .github/workflows or .gitlab-ci.yml pushed without reviewSCM Git and push eventsMain route for poisoned pipeline execution (ATT&CK T1677)
Protection rule, ruleset or branch policy removed or bypassedSCM audit logRemoves flow control before a malicious change
Bulk repository clone or archive download by one tokenSCM Git events or API request eventsCode collection (ATT&CK T1213.003 Code Repositories)
Runner making outbound connections to unknown hosts during a buildRunner network or DNS logsSecret exfiltration from build jobs
Artifact or image pushed by an identity other than the CI service accountRegistry audit logArtifact tampering (see CICD-SEC-9)
Audit stream paused, deleted or retargetedSCM or platform audit logDefence impairment before an attack

Evidence and assessment

An assessor will want to see that logging is complete, protected, retained and acted on. Prepare evidence for each of these:

  • Inventory and coverage: the CI/CD system inventory with log status per instance, and the percentage of instances streaming to the central store.
  • Configuration: screenshots or API exports showing audit logging enabled (for example "Log Audit Events" in Azure DevOps), stream destinations, API request events, source IP disclosure and the Kubernetes audit policy file.
  • Retention: the retention set in the central store against policy. NIST SP 800-53 AU-11 asks for a defined retention period; native platform retention alone (90 or 180 days) rarely meets it.
  • Integrity: access controls on log storage and an alert record for stream or setting changes (AU-9).
  • Use: the detection rule list for CI/CD sources, test results from exercises, and closed tickets showing alerts were triaged.

Metrics that show the control is working:

  • Coverage: share of inventoried CI/CD instances sending both audit and applicative logs (target 100 percent of production-relevant systems).
  • Attribution: share of pipeline events with a resolvable identity type and owner.
  • Latency: median time from event to availability in the SIEM.
  • Detection test pass rate: share of exercised scenarios that raised an alert.
  • Time to detect: median time from a pipeline security event to triage, tracked across exercises and real incidents.

Mapping to other frameworks

FrameworkReferenceRelevance to CICD-SEC-10
NIST SP 800-53 Rev. 5AU-2 Event Logging, AU-3 Content of Audit Records, AU-12 Audit Record GenerationDecide and generate the events, with actor and context
NIST SP 800-53 Rev. 5AU-6 Audit Record Review, Analysis, and Reporting; SI-4 System MonitoringAlerts and review of CI/CD activity
NIST SP 800-53 Rev. 5AU-9 Protection of Audit Information; AU-11 Audit Record RetentionTamper protection and retention beyond platform defaults
NIST SP 800-53 Rev. 5CM-8 System Component Inventory; IR-4 Incident HandlingMapping the environment; investigation support
NIST CSF 2.0ID.AM-02, PR.PS-04, DE.CM-09, DE.AE-03, RS.AN-03Inventory, log generation, monitoring of runtime environments, correlation, incident analysis
CIS Controls v8.1Control 8 Audit Log ManagementCollection, centralisation, retention and review of audit logs
ISO/IEC 27001:2022 Annex AA.8.15 Logging; A.8.16 Monitoring activitiesLogging and monitoring controls applied to engineering systems
MITRE ATT&CK v20T1677 Poisoned Pipeline Execution; T1195.002 Compromise Software Supply Chain; T1685 Disable or Modify ToolsTechniques the logs must make visible

See the CIS Control 8 Audit Log Management Guide for the general logging programme this guide applies to CI/CD.

Common pitfalls

  • Relying on the platform's own console. Native audit logs expire (90 days in Azure DevOps, 180 in a GitHub organisation). The CircleCI breach shows why time matters: the first signal came from a customer 13 days after the initial compromise, and investigators still had to reconstruct what happened before that.
  • Logging humans, not tokens. In the GhostAction campaign, malicious workflows were pushed under maintainers' own accounts and read secrets the existing pipelines already used. Detection came from an outside researcher, not from the victims' own monitoring of workflow changes or runner egress.
  • Forgetting self-hosted instances. A Jenkins controller set up by one team, without the Audit Trail plugin, is invisible. OWASP's call to inventory every instance exists for this reason.
  • Audit logs without applicative logs. Permission changes alone will not show a build that ran an unexpected step or an artifact uploaded from outside CI. OWASP gives both equal importance.
  • Leaving defaults in place. Hidden source IPs, API request events not streamed and auditing switched off all look like coverage on an architecture diagram.
  • Capturing secrets in logs. Verbose build output or Kubernetes RequestResponse logging of secrets can put credentials into the SIEM. Mask outputs and scope verbose levels carefully. See the Secure Error Handling and Logging Guide.

Practitioner checklist

  • Maintain an inventory of every CI/CD system instance, with an owner and its log status.
  • Enable audit logging everywhere it is off by default, including Azure DevOps "Log Audit Events" and GitHub source IP disclosure.
  • Stream SCM audit and Git events, CI logs, registry and cloud audit trails to a central store with retention set by policy, not by platform default.
  • Tag every event with an identity type and map pipeline identities to owners.
  • Log credential use by jobs and join cloud role assumptions to pipeline run IDs.
  • Protect log storage from pipeline identities and alert on any stream or audit setting change.
  • Deploy detections for workflow file changes, protection removal, new tokens and keys, runner registration and unexpected artifact publishers.
  • Add at least one cross-system rule linking commit, run and artifact.
  • Exercise the detections at least twice a year and track coverage, latency and time to detect.

Standards and references

Related NHI Mgmt Group resources: Top 10 CI/CD Security Risks · CI/CD Pipeline Identity Security Guide · GitHub and GitHub Actions Security Guide · SIEM Guide · CI/CD Pipeline Compromise Playbook

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 26 October 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.