Module 5: Integrate AIRS into Pipeline

Overview

This is the "customer engagement" moment. Your pipeline works. Models train, merge, publish, and deploy. But right now, any model artifact -- clean or compromised -- flows straight to production without a single security check.

You will add AIRS scanning to Gate 2, manifest verification to Gate 3, metadata labeling for traceability, and detailed evaluation reporting directly in GitHub Actions.

Interactive Lab

The full interactive experience for this module runs in Claude Code. Use /lab:module 5 to begin the guided walkthrough with your AI mentor.

Objectives

• Add an AIRS scan step to Gate 2 between merge and publish
• Add manifest provenance verification to Gate 3 before deployment
• Implement scan labeling with pipeline metadata (gate, run ID, model version, environment, base model)
• Pull per-rule evaluation details and render them in GitHub Actions summaries

Time Estimate

~1 to 1.5 hours

Challenges

5.1: Add Scanning to Gate 2

Gate 2 currently merges and publishes without any security validation. Add an AIRS scan between merge and publish. If the scan returns BLOCKED, the pipeline must halt -- the compromised model never reaches the production registry.

What to modify: .github/workflows/gate-2-publish.yaml

5.2: Add Manifest Verification to Gate 3

Gate 3 currently deploys whatever model path it receives without checking whether that model was ever scanned. Add manifest verification that requires a Gate 2 scan record before allowing deployment.

The verification tool: scripts/manifest.py verify --require-scan gate2

5.3: Label Your Scans

Add meaningful key-value labels to scan requests so they are identifiable and queryable in SCM. Labels like gate, run_id, model_version, environment, and base_model create traceability from any scan back to the exact pipeline run.

5.4: Enrich Scan Output for Developers

Surface scan results in the GitHub Actions job summary so developers see exactly what happened without leaving GitHub. Build a rich markdown summary with verdict, rule counts, scan UUID, and labels.

Product gap to discuss: Per-rule evaluation details are not available through the SDK. Full breakdowns require SCM access. This is a real limitation worth raising with customers.

Key Concepts

• Pipeline Scanning -- The enforcement point. Without it, any artifact flows to production unchecked
• Manifest Verification -- Chain of custody from training to deployment. Has this exact artifact been scanned?
• Scan Labels -- Connect scans to business context. Make "which models passed scanning this quarter?" answerable
• GitHub Actions Summaries -- Rich markdown rendered in the workflow run UI via $GITHUB_STEP_SUMMARY

Verification

Run /lab:verify-5 in Claude Code to confirm scanning in Gate 2, manifest verification in Gate 3, labels visible in SCM, and rich summaries in GitHub Actions.

What's Next

Your pipeline is secured. But what exactly is AIRS catching? Module 6: The Threat Zoo builds malicious models, scans them, and explores the real threat landscape.