DEPLOY
Owner: DevOps / Platform Engineering / Change Advisory Board
Controlled AI Deployment & Release
Manage safe, versioned, and auditable deployment of AI models and agent systems with rollback capabilities and environment isolation.
Framework Mapping
Controls from each source framework that map to this domain.
| Framework | Mapped Controls |
|---|---|
| ISO 42001 |
A.5 Assessing AI System Impact (deploy)
A.7 AI System Operation
|
| NIST AI RMF |
MG-1 Allocate Risk Mgmt
MG-2 Map & Measure
MG-3 Manage AI Risks
MG-4 Risk Treatment
|
| OWASP |
ASI04 Agentic Supply Chain
ASI08 Compliance & Regulatory
|
Controls
7 controls across Tier 1 (essential) and Tier 2 (advanced).
Tier 1
ISO A.5
NIST MG-1
Deployment Readiness Gate
Tier 1
NIST MG-2
Infrastructure Hardening
Tier 1
ISO A.7
NIST MG-3
Model Versioning & Rollback
Tier 2
NIST MG-4
Canary/Blue-Green Deployment
Tier 1
Environment Isolation
Tier 2
ISO A.5
NIST MG-3
Change Management for Models
Tier 2
NIST MG-1
SLA & Performance Baselines
Audit Checklist
Quick-reference checklist items grouped by control.
- ☐ Readiness gate policy exists defining criteria and approval authority
- ☐ Recent deployments have completed checklists on file
- ☐ Review meetings conducted with documented attendance and decisions
- ☐ Approval records match required authority level for system risk tier
- ☐ Post-deployment validation completed within 1 week of launch
- ☐ Patch management process defined with monthly cadence and evidence of recent patches
- ☐ IAM policies follow least privilege with quarterly review records
- ☐ Encryption at rest and in transit enabled and validated
- ☐ Unnecessary services disabled with security group/firewall rules enforced
- ☐ Vulnerability scans show <5 critical/high findings with remediation plan
- ☐ Versioning scheme defined and applied to all production models
- ☐ Model registry maintains version history with metadata
- ☐ Rollback procedure documented and automated where possible
- ☐ Rollback tested quarterly with <15 minute RTO demonstrated
- ☐ Artifact retention policy enforced with evidence of archival
- ☐ Canary or blue-green strategy defined and documented
- ☐ Success metrics and rollback thresholds configured
- ☐ Traffic routing implemented and tested (gradual rollout)
- ☐ Recent deployments show canary pattern with monitoring logs
- ☐ Auto-rollback tested and validated within last quarter
- ☐ Environments deployed in separate networks with ACLs enforced
- ☐ Credentials unique per environment, no shared secrets
- ☐ Production access requires MFA and VPN/bastion
- ☐ Production data prohibited in non-prod with masking/anonymization enforced
- ☐ Access logs reviewed quarterly showing compliance with access policies
- ☐ Change management process defined with ticket requirements
- ☐ Recent changes have tickets with all required fields completed
- ☐ Peer review and approval documented for all changes
- ☐ Change log maintained with entries for all production changes
- ☐ Emergency change process defined with post-facto approval records
- ☐ SLAs defined for all production AI systems
- ☐ Baselines established using production data or load tests
- ☐ Monitoring configured with alerts for SLA violations
- ☐ Alert history shows timely response to SLA violations
- ☐ Quarterly reviews conducted with baseline adjustments documented