Third-Party AI Risk Management Standard

Document Type: Enterprise Standard
Framework Alignment: NIST AI RMF 2.0, EU AI Act, ISO/IEC 42001 Annex A.5
Effective Date: 2026-01-15
Version: 1.0

1. Purpose and Scope

1.1 Purpose

This standard establishes requirements for managing risks associated with third-party AI components, services, and vendors. It ensures that external AI dependencies receive appropriate due diligence, monitoring, and governance throughout their lifecycle.

1.2 Scope

This standard applies to:

Third-party AI APIs (OpenAI, Anthropic, Google, etc.)
Embedded AI features in SaaS products
Open-source AI models and frameworks
AI consulting and development services
Data labeling and annotation services
AI infrastructure providers (GPU cloud, model serving)

1.3 Regulatory Drivers

Regulation	Requirement	Reference
NIST AI RMF 2.0	Third-party model assessment	MAP 1.5, MANAGE 2.3
EU AI Act	Value chain responsibilities	Article 25
EU AI Act	GPAI downstream obligations	Article 53(1)(b)
ISO/IEC 42001	Third-party management	Annex A.5.4

2. Third-Party AI Classification

2.1 Classification Categories

Category	Definition	Examples	Risk Level
API Services	AI capabilities accessed via API	OpenAI, Anthropic, Google AI	High
Embedded AI	AI features within SaaS products	Salesforce Einstein, ServiceNow	Medium-High
Open Source Models	Community-developed AI models	Llama, Mistral, Stable Diffusion	Medium
AI Infrastructure	Computing and serving platforms	AWS Bedrock, Azure AI	Medium
Data Services	Training data and labeling	Scale AI, Labelbox	Medium-High
Development Services	AI consulting and development	Custom development vendors	High

2.2 Risk Tier Assignment

Third-Party Risk Tier	Criteria	Due Diligence Level
Critical	Core business dependency; handles sensitive data; GPAI provider	Full assessment + continuous monitoring
High	Significant business function; some sensitive data	Enhanced assessment + periodic monitoring
Medium	Supporting function; limited data exposure	Standard assessment + annual review
Low	Minimal impact; no sensitive data	Simplified assessment

3. Due Diligence Requirements

3.1 Pre-Engagement Assessment

3.1.1 Vendor Assessment Questionnaire

Section A: Company Information

Legal entity name and jurisdiction
Company size and financial stability
AI-specific certifications (ISO/IEC 42001, SOC 2)
Insurance coverage (AI/ML specific)

Section B: AI Governance

AI ethics policy and governance structure
Responsible AI practices documentation
Incident response procedures
Model change management process

Section C: Technical Capabilities

Model documentation availability
API versioning and deprecation policy
SLA and uptime commitments
Scalability and performance guarantees

Section D: Security

Security certifications (SOC 2, ISO 27001)
Data encryption (at rest, in transit)
Access control mechanisms
Penetration testing frequency

Section E: Data Handling

Data processing locations
Data retention policies
Training data usage (opt-out available?)
Sub-processor disclosure

Section F: Compliance

EU AI Act compliance status
GDPR/privacy compliance
Industry-specific certifications
Audit rights provision

3.2 Assessment Scoring Matrix

Category	Weight	Score (1-5)	Weighted Score
AI Governance	20%
Security Posture	25%
Data Handling	20%
Regulatory Compliance	20%
Technical Capability	15%
Total	100%

Scoring Thresholds:

≥4.0: Approved for all tiers
3.0-3.9: Approved with conditions
2.0-2.9: Enhanced monitoring required
<2.0: Not approved

3.3 Documentation Requirements by Tier

Document	Critical	High	Medium	Low
Vendor Assessment	✓ Full	✓ Full	✓ Standard	✓ Simplified
AI-BOM Entry	✓	✓	✓	✓
Security Review	✓	✓	Optional	-
Legal Review	✓	✓	Optional	-
Privacy Impact Assessment	✓	If PII	-	-
Model Documentation	✓	✓	Summary	-
Training Data Summary	✓ (if GPAI)	✓ (if GPAI)	-	-

4. Contract Requirements

4.1 Mandatory Contract Clauses

All Third-Party AI Contracts Must Include:

Clause	Purpose	Minimum Standard
AI Disclosure	Transparency	Vendor must disclose all AI components
Data Usage	Data protection	No use of our data for model training without explicit consent
IP Indemnification	Legal protection	Vendor indemnifies for IP infringement claims
Audit Rights	Assurance	Right to audit or receive audit reports
Incident Notification	Risk management	24-48 hour notification for AI incidents
Subprocessor Notification	Supply chain	Notification of AI subprocessor changes
Documentation Access	Compliance	Access to model documentation on request
Exit Provisions	Business continuity	Data portability, transition assistance

4.2 EU AI Act Specific Clauses

For GPAI Providers:

Obligation to provide downstream documentation (Article 53(1)(b))
Training data summary access
Systemic risk notification commitment
Compliance attestation

For High-Risk AI Components:

Conformity assessment documentation
Technical documentation access
Human oversight support
Traceability requirements

4.3 SLA Requirements

Metric	Critical Tier	High Tier	Medium Tier
Uptime	99.9%	99.5%	99.0%
Response Time (P95)	<200ms	<500ms	<1000ms
Incident Response	1 hour	4 hours	24 hours
Model Update Notice	30 days	14 days	7 days
Support Hours	24/7	Business hours + on-call	Business hours

5. Ongoing Monitoring

5.1 Continuous Monitoring Requirements

Monitoring Area	Frequency	Method	Owner
Service Availability	Real-time	API monitoring	Platform Team
Performance Metrics	Daily	SLA dashboard	Platform Team
Security Incidents	Continuous	Vendor notifications + news	Security Team
Model Changes	On notification	Change review	AI Governance
Compliance Updates	Quarterly	Vendor attestation	Compliance
Financial Stability	Annual	Credit check, news	Procurement

5.2 Model Drift and Performance Monitoring

For AI API Services:

Monitor output quality metrics
Track response time trends
Log and analyze error rates
Compare outputs against baselines

Monitoring Dashboard Requirements:

API call volumes and costs
Error rates by endpoint
Latency distribution
Quality score trends

5.3 Risk Event Triggers

Event	Risk Level	Action Required
Vendor security breach	Critical	Immediate assessment, potential suspension
Model deprecation notice	High	Migration planning within 30 days
Terms of service change	Medium	Legal review within 14 days
Pricing change	Medium	Cost impact analysis
Acquisition/merger	High	Vendor reassessment
Regulatory enforcement action	Critical	Compliance review, potential exit

6. Vendor Management Lifecycle

6.1 Lifecycle Stages

┌─────────────────────────────────────────────────────────────┐
│ 1. IDENTIFICATION                                           │
│    • Business need definition                               │
│    • Market research                                        │
│    • Shortlist candidates                                   │
└─────────────────────────────────────────────────────────────┘
                              │
                              ▼
┌─────────────────────────────────────────────────────────────┐
│ 2. ASSESSMENT                                               │
│    • Vendor questionnaire                                   │
│    • Security review                                        │
│    • Legal review                                           │
│    • Technical evaluation                                   │
└─────────────────────────────────────────────────────────────┘
                              │
                              ▼
┌─────────────────────────────────────────────────────────────┐
│ 3. CONTRACTING                                              │
│    • Contract negotiation                                   │
│    • Required clauses inclusion                             │
│    • SLA definition                                         │
│    • Approval workflow                                      │
└─────────────────────────────────────────────────────────────┘
                              │
                              ▼
┌─────────────────────────────────────────────────────────────┐
│ 4. ONBOARDING                                               │
│    • Technical integration                                  │
│    • AI-BOM creation                                        │
│    • Monitoring setup                                       │
│    • Team training                                          │
└─────────────────────────────────────────────────────────────┘
                              │
                              ▼
┌─────────────────────────────────────────────────────────────┐
│ 5. ONGOING MANAGEMENT                                       │
│    • Continuous monitoring                                  │
│    • Periodic reassessment                                  │
│    • Incident management                                    │
│    • Relationship management                                │
└─────────────────────────────────────────────────────────────┘
                              │
                              ▼
┌─────────────────────────────────────────────────────────────┐
│ 6. OFFBOARDING                                              │
│    • Exit planning                                          │
│    • Data retrieval/deletion                                │
│    • Alternative sourcing                                   │
│    • Documentation archive                                  │
└─────────────────────────────────────────────────────────────┘

6.2 Reassessment Schedule

Tier	Full Reassessment	Security Review	Compliance Check
Critical	Annual	Quarterly	Quarterly
High	Annual	Semi-annual	Semi-annual
Medium	Biennial	Annual	Annual
Low	On renewal	On renewal	On renewal

7. Open Source AI Model Requirements

7.1 Open Source Assessment Criteria

Criterion	Assessment	Minimum Standard
License Compliance	Legal review	Compatible with commercial use
Provenance	Documentation	Clear model lineage
Security	Vulnerability scan	No known critical CVEs
Community Support	Activity assessment	Active maintenance
Training Data	Transparency	Known data sources

7.2 Open Source AI Checklist

Before Adoption:

License reviewed and approved by Legal
Model provenance documented
Security scan completed (no critical vulnerabilities)
Hash/checksum verified against source
Training data sources understood
AI-BOM entry created
Maintenance responsibility assigned

Ongoing:

Monitor for security advisories
Track version updates
Maintain local copies with verified hashes
Periodic re-evaluation (annual minimum)

7.3 Model Download and Verification

# Example: Model verification procedure
# 1. Download model from official source
wget https://official-source/model-v1.0.safetensors

# 2. Verify hash against published checksum
sha256sum model-v1.0.safetensors
# Compare output with official checksum

# 3. Document in AI-BOM
# Model: model-v1.0
# Source: official-source
# Download Date: YYYY-MM-DD
# SHA-256: [verified hash]

8. Incident Response

8.1 Third-Party AI Incident Types

Incident Type	Severity	Response Time
Vendor data breach involving our data	Critical	1 hour
Model producing harmful outputs	High	4 hours
Service outage (critical dependency)	High	1 hour
Model significant behavior change	Medium	24 hours
Pricing/terms change	Low	7 days

8.2 Incident Response Procedure

Immediate (0-4 hours):

Assess impact on our systems
Activate alternative/backup if available
Notify internal stakeholders
Document incident details

Short-term (4-48 hours):

Obtain vendor incident report
Assess ongoing risk
Implement compensating controls
Communication to affected parties

Long-term (post-incident):

Conduct root cause analysis
Update vendor risk assessment
Review contract adequacy
Implement preventive measures

9. SMB Implementation Guide

9.1 Simplified Assessment

For SMBs, a simplified vendor assessment approach:

Tier 1 Check (5 minutes):

Vendor has privacy policy
Vendor has security page/certifications
Terms of service reviewed
Pricing understood

Tier 2 Check (30 minutes):

Security certifications verified (SOC 2, ISO 27001)
Data processing agreement in place
API documentation reviewed
Support availability confirmed

Tier 3 Check (2-4 hours):

Full vendor questionnaire
Legal review of contract
Technical integration assessment
Reference checks

9.2 SMB Vendor Tracking

Simple spreadsheet approach:

Vendor	Service	Tier	Contract End	Last Review	Owner

10. Enterprise Integration

10.1 GRC Integration

Third-party AI risk management should integrate with:

Vendor Risk Management (VRM) system
GRC platform (ServiceNow, Archer, OneTrust)
Contract management system
Asset management / CMDB

10.2 Process Integration

Process	Third-Party AI Touchpoint
Procurement	Vendor assessment before PO
Security Review	AI-specific security questionnaire
Legal Review	AI contract clause checklist
Architecture Review	AI-BOM requirements
Vendor Review Board	AI risk tier consideration

11. Metrics and Reporting

11.1 Key Metrics

Metric	Target	Frequency
% vendors with current assessment	100%	Monthly
Average vendor risk score	≥3.5	Quarterly
Overdue reassessments	0	Monthly
Incidents from third-party AI	Minimize	Monthly
Contract compliance rate	100%	Quarterly

11.2 Reporting Requirements

Monthly Report:

New third-party AI onboarded
Vendor risk score changes
Incidents and resolutions
Upcoming reassessments

Quarterly Report:

Third-party AI inventory summary
Risk distribution by tier
Compliance status
Cost trends

Document Control

Version History

Version	Date	Author	Changes
1.0	2025-06-15	AI Governance Office	Initial release

Approvals

Role	Name	Date
AI Risk Officer
Procurement Lead
CISO
Legal Counsel

Appendix A: Vendor Assessment Questionnaire Template

[Full questionnaire available as separate attachment]

Appendix B: Contract Clause Library

[Standard AI contract clauses available as separate attachment]

Appendix C: Approved Vendor List

[Internal list maintained separately with access controls]

CODITECT AI Risk Management Framework

Document ID: AI-RMF-15 | Version: 2.0.0 | Status: Active

AZ1.AI Inc. | CODITECT Platform

Framework Alignment: NIST AI RMF 2.0 | EU AI Act | ISO/IEC 42001

This document is part of the CODITECT AI Risk Management Framework. For questions or updates, contact the AI Governance Office.

Repository: coditect-ai-risk-management-framework Last Updated: 2026-01-15 Owner: AZ1.AI Inc. | Lead: Hal Casteel

1. Purpose and Scope​

1.1 Purpose​

1.2 Scope​

1.3 Regulatory Drivers​

2. Third-Party AI Classification​

2.1 Classification Categories​

2.2 Risk Tier Assignment​

3. Due Diligence Requirements​

3.1 Pre-Engagement Assessment​

3.1.1 Vendor Assessment Questionnaire​

3.2 Assessment Scoring Matrix​

3.3 Documentation Requirements by Tier​

4. Contract Requirements​

4.1 Mandatory Contract Clauses​

4.2 EU AI Act Specific Clauses​

4.3 SLA Requirements​

5. Ongoing Monitoring​

5.1 Continuous Monitoring Requirements​

5.2 Model Drift and Performance Monitoring​

5.3 Risk Event Triggers​

6. Vendor Management Lifecycle​

6.1 Lifecycle Stages​

6.2 Reassessment Schedule​

7. Open Source AI Model Requirements​

7.1 Open Source Assessment Criteria​

7.2 Open Source AI Checklist​

7.3 Model Download and Verification​

8. Incident Response​

8.1 Third-Party AI Incident Types​

8.2 Incident Response Procedure​

9. SMB Implementation Guide​

9.1 Simplified Assessment​

9.2 SMB Vendor Tracking​

10. Enterprise Integration​

10.1 GRC Integration​

10.2 Process Integration​

11. Metrics and Reporting​

11.1 Key Metrics​

11.2 Reporting Requirements​

Document Control​

Version History​

Approvals​

Appendix A: Vendor Assessment Questionnaire Template​

Appendix B: Contract Clause Library​

Appendix C: Approved Vendor List​