What this control covers

A controlled AI system design and development process shall be established, implemented, and continually improved.

Why it matters

Vulnerabilities introduced in code are 100× cheaper to fix in development than in production; this control is the place where security shifts left.

How to implement

AWS

Use CodeBuild with VPC-isolated build environments and CodeArtifact for an internal package mirror so builds don't pull straight from the public registry. Sign artefacts with AWS Signer.

Azure

Use Azure DevOps Pipelines with mandatory approvals on production environments and Microsoft Defender for DevOps for secret scanning across GitHub/ADO/GitLab in one pane.

GCP

Vertex AI Model Registry with model-card metadata, Model Garden for vetted base models, and Sensitive Data Protection on training/inference inputs.

GitHub

Apply branch protection on default and release branches: require PR review, status checks, signed commits, linear history, and dismiss stale reviews on new commits. Use CODEOWNERS for required reviewers, Dependabot + CodeQL for SAST, and the Dependency graph / SBOM export for vendor disclosures.

Evidence to collect

  • policy_document — A signed, version-controlled policy stating leadership intent and the named owner.
  • code_repo_scan — See the iCompaas Evidence library for the canonical source for this evidence type.
  • ticketing_workflow — See the iCompaas Evidence library for the canonical source for this evidence type.

Policies, repository controls, and workflow evidence should demonstrate an established and controlled AI design and development process, including reviews, approvals, and continuous improvement activities. Focus: AI system design and development process.

Quick checklist

  • Develop — Development Process

Related articles on this site